DPDK: update to DPDK 2.1

- support both DPDK 2.0 and 2.1 - use mbuf priv_size for vr_packet - use new packet type flags - use rte_pktmbuf_mtod_offset() - count out HW XEC counter from ierrors - use RTE_PORT_STATS_COLLECT Change-Id: I2015961e95bb02a78b7419f7098a1e48bd622356 Closes-bug: 1430629
Juniper · Sep 18, 2015 · 72f6ca0 · 72f6ca0
1 parent 3dcdb0d
commit 72f6ca0
Show file tree

Hide file tree

Showing 9 changed files with 135 additions and 25 deletions.
diff --git a/dpdk/dpdk_vrouter.c b/dpdk/dpdk_vrouter.c
@@ -65,21 +65,43 @@ static int dpdk_argc = RTE_DIM(dpdk_argv) - 10;
 /* Timestamp logger */
 static FILE *timestamp_log_stream;
 
-/* Pktmbuf constructor with vr_packet support */
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+/* A packet mbuf pool constructor with vr_packet support */
+void vr_dpdk_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg)
+{
+    struct rte_pktmbuf_pool_private priv;
+
+    /* Set private mbuf size for vr_packet. */
+    priv.mbuf_data_room_size = mp->elt_size - sizeof(struct rte_mbuf)
+        - sizeof(struct vr_packet);
+    priv.mbuf_priv_size = sizeof(struct vr_packet);
+
+    rte_pktmbuf_pool_init(mp, &priv);
+}
+#else
+void vr_dpdk_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg)
+{
+    rte_pktmbuf_pool_init(mp, opaque_arg);
+}
+#endif
+
+/* The packet mbuf constructor with vr_packet support */
 void
 vr_dpdk_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg, void *_m, unsigned i)
 {
     struct rte_mbuf *m = _m;
     struct vr_packet *pkt;
     rte_pktmbuf_init(mp, opaque_arg, _m, i);
 
+#if (RTE_VERSION < RTE_VERSION_NUM(2, 1, 0, 0))
     /* decrease rte packet size to fit vr_packet struct */
     m->buf_len -= sizeof(struct vr_packet);
     RTE_VERIFY(0 < m->buf_len);
 
     /* start of buffer is just after vr_packet structure */
     m->buf_addr += sizeof(struct vr_packet);
     m->buf_physaddr += sizeof(struct vr_packet);
+#endif
 
     /* basic vr_packet initialization */
     pkt = vr_dpdk_mbuf_to_pkt(m);
@@ -95,7 +117,7 @@ dpdk_mempools_create(void)
     vr_dpdk.rss_mempool = rte_mempool_create("rss_mempool", VR_DPDK_RSS_MEMPOOL_SZ,
             VR_DPDK_MBUF_SZ, VR_DPDK_RSS_MEMPOOL_CACHE_SZ,
             sizeof(struct rte_pktmbuf_pool_private),
-            rte_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
+            vr_dpdk_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
             rte_socket_id(), 0);
     if (vr_dpdk.rss_mempool == NULL) {
         RTE_LOG(CRIT, VROUTER, "Error creating RSS mempool: %s (%d)\n",
@@ -107,7 +129,7 @@ dpdk_mempools_create(void)
     vr_dpdk.frag_direct_mempool = rte_mempool_create("frag_direct_mempool",
             VR_DPDK_FRAG_DIRECT_MEMPOOL_SZ, VR_DPDK_FRAG_DIRECT_MBUF_SZ,
             VR_DPDK_FRAG_DIRECT_MEMPOOL_CACHE_SZ,
-            sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init,
+            sizeof(struct rte_pktmbuf_pool_private), vr_dpdk_pktmbuf_pool_init,
             NULL, vr_dpdk_pktmbuf_init, NULL, rte_socket_id(), 0);
     if (vr_dpdk.frag_direct_mempool == NULL) {
         RTE_LOG(CRIT, VROUTER, "Error creating FRAG_DIRECT mempool: %s (%d)\n",
@@ -141,7 +163,7 @@ dpdk_mempools_create(void)
         vr_dpdk.free_mempools[i] = rte_mempool_create(mempool_name,
                 VR_DPDK_VM_MEMPOOL_SZ, VR_DPDK_MBUF_SZ, VR_DPDK_VM_MEMPOOL_CACHE_SZ,
                 sizeof(struct rte_pktmbuf_pool_private),
-                rte_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
+                vr_dpdk_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
                 rte_socket_id(), 0);
         if (vr_dpdk.free_mempools[i] == NULL) {
             RTE_LOG(CRIT, VROUTER, "Error creating VM mempool %d: %s (%d)\n",

diff --git a/dpdk/vr_dpdk_ethdev.c b/dpdk/vr_dpdk_ethdev.c
@@ -704,7 +704,9 @@ dpdk_mbuf_rss_hash(struct rte_mbuf *mbuf)
     /* TODO: IPv6 support */
     /* TODO: VLAN support */
     if (likely(eth_hdr->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))) {
-        mbuf->ol_flags |= PKT_RX_IPV4_HDR;
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+        mbuf->packet_type |= RTE_PTYPE_L3_IPV4;
+#endif
         ipv4_hdr = (struct ipv4_hdr *)((uintptr_t)eth_hdr
                                     + sizeof(struct ether_hdr));
         ipv4_addr_ptr = (uint64_t *)((uintptr_t)ipv4_hdr
@@ -723,8 +725,17 @@ dpdk_mbuf_rss_hash(struct rte_mbuf *mbuf)
                     !vr_ip_fragment((struct vr_ip *)ipv4_hdr))) {
             switch (ipv4_hdr->next_proto_id) {
             case IPPROTO_TCP:
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+                mbuf->packet_type |= RTE_PTYPE_L4_TCP;
+#endif
+                l4_ptr = (uint32_t *)((uintptr_t)ipv4_hdr + iph_len);
+
+                hash = rte_hash_crc_4byte(*l4_ptr, hash);
+                break;
             case IPPROTO_UDP:
-                mbuf->ol_flags |= PKT_RX_IPV4_HDR_EXT;
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+                mbuf->packet_type |= RTE_PTYPE_L4_UDP;
+#endif
                 l4_ptr = (uint32_t *)((uintptr_t)ipv4_hdr + iph_len);
 
                 hash = rte_hash_crc_4byte(*l4_ptr, hash);
@@ -754,8 +765,8 @@ vr_dpdk_ethdev_rx_emulate(struct vr_interface *vif, struct rte_mbuf *pkts[VR_DPD
 
     /* prefetch the mbufs */
     for (i = 0; i < nb_pkts; i++) {
-        rte_prefetch0(rte_pktmbuf_mtod(pkts[i], void *));
-        rte_prefetch0(rte_pktmbuf_mtod(pkts[i], uint8_t *) + RTE_CACHE_LINE_SIZE);
+        rte_prefetch0(rte_pktmbuf_mtod(pkts[i], uint8_t *));
+        rte_prefetch0(rte_pktmbuf_mtod_offset(pkts[i], uint8_t *, RTE_CACHE_LINE_SIZE));
     }
 
     /* emulate VLAN stripping if needed */

diff --git a/dpdk/vr_dpdk_host.c b/dpdk/vr_dpdk_host.c
@@ -323,7 +323,7 @@ dpdk_pktmbuf_copy_bits(const struct rte_mbuf *mbuf, int offset,
     do {
         if (offset < rte_pktmbuf_data_len(mbuf)) {
             /* copy a piece of data */
-            from = (void *)(rte_pktmbuf_mtod(mbuf, uintptr_t) + offset);
+            from = (void *)(rte_pktmbuf_mtod_offset(mbuf, uintptr_t, offset));
             copy = rte_pktmbuf_data_len(mbuf) - offset;
             if (copy > len)
                 copy = len;
@@ -640,7 +640,7 @@ dpdk_pheader_pointer(struct vr_packet *pkt, unsigned short hdr_len, void *buf)
         int len = rte_pktmbuf_data_len(m) - offset;
         void *tmp_buf = buf;
 
-        rte_memcpy(tmp_buf, rte_pktmbuf_mtod(m, char *) + offset, len);
+        rte_memcpy(tmp_buf, rte_pktmbuf_mtod_offset(m, char *, offset), len);
         hdr_len -= len;
         tmp_buf = (void *)((uintptr_t)tmp_buf + len);
 

diff --git a/dpdk/vr_dpdk_interface.c b/dpdk/vr_dpdk_interface.c
@@ -1303,6 +1303,35 @@ dpdk_dev_stats_update(struct vr_interface *vif, unsigned lcore_id)
     if (rte_eth_stats_get(port_id, &eth_stats) != 0)
         return;
 
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+    /*
+     * TODO: In DPDK 2.1 ierrors includes XEC (l3_l4_xsum_error) counter.
+     * The counter seems to include no check sum UDP packets. As a workaround
+     * we count out the XEC from ierrors using rte_eth_xstats_get().
+     */
+    struct rte_eth_xstats *eth_xstats = NULL;
+    int nb_xstats, i;
+    nb_xstats = rte_eth_xstats_get(port_id, eth_xstats, 0);
+    if (nb_xstats > 0) {
+        eth_xstats = rte_malloc("xstats",
+            sizeof(*eth_xstats)*nb_xstats, 0);
+        if (eth_xstats != NULL) {
+            if (rte_eth_xstats_get(port_id, eth_xstats, nb_xstats)
+                    == nb_xstats) {
+                /* look for XEC counter */
+                for (i = 0; i < nb_xstats; i++) {
+                    if (strncmp(eth_xstats[i].name, "l3_l4_xsum_error",
+                        sizeof(eth_xstats[i].name)) == 0) {
+                        eth_stats.ierrors -= eth_xstats[i].value;
+                        break;
+                    }
+                }
+            }
+            rte_free(eth_xstats);
+        }
+    }
+#endif
+
     /* per-lcore device counters */
     lcore = vr_dpdk.lcores[lcore_id];
     if (lcore == NULL)

diff --git a/dpdk/vr_dpdk_knidev.c b/dpdk/vr_dpdk_knidev.c
@@ -24,7 +24,7 @@
 /*
  * KNI Reader
  */
-#if DPDK_KNIDEV_READER_STATS_COLLECT == 1
+#ifdef RTE_PORT_STATS_COLLECT
 
 #define DPDK_KNIDEV_READER_STATS_PKTS_IN_ADD(port, val) \
     port->stats.n_pkts_in += val
@@ -121,7 +121,7 @@ dpdk_knidev_reader_stats_read(void *port,
 /*
  * KNI Writer
  */
-#if DPDK_KNIDEV_WRITER_STATS_COLLECT == 1
+#ifdef RTE_PORT_STATS_COLLECT
 
 #define DPDK_KNIDEV_WRITER_STATS_PKTS_IN_ADD(port, val) \
     port->stats.n_pkts_in += val

diff --git a/dpdk/vr_dpdk_usocket.c b/dpdk/vr_dpdk_usocket.c
@@ -785,7 +785,7 @@ usock_alloc(unsigned short proto, unsigned short type)
             usockp->usock_mbuf_pool = rte_mempool_create("packet_mbuf_pool",
                     PKT0_MBUF_POOL_SIZE, PKT0_MBUF_PACKET_SIZE,
                     PKT0_MBUF_POOL_CACHE_SZ, sizeof(struct rte_pktmbuf_pool_private),
-                    rte_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
+                    vr_dpdk_pktmbuf_pool_init, NULL, vr_dpdk_pktmbuf_init, NULL,
                     rte_socket_id(), 0);
             if (!usockp->usock_mbuf_pool)
                 goto error_exit;

diff --git a/dpdk/vr_dpdk_virtio.c b/dpdk/vr_dpdk_virtio.c
@@ -469,7 +469,7 @@ vr_dpdk_guest_phys_to_host_virt(vr_uvh_client_t *vru_cl, uint64_t paddr)
     return NULL;
 }
 
-#if DPDK_VIRTIO_READER_STATS_COLLECT == 1
+#ifdef RTE_PORT_STATS_COLLECT
 
 #define DPDK_VIRTIO_READER_STATS_PKTS_IN_ADD(port, val) \
         port->stats.n_pkts_in += val
@@ -626,7 +626,7 @@ dpdk_virtio_from_vm_rx(void *port, struct rte_mbuf **pkts, uint32_t max_pkts)
     return pkts_sent;
 }
 
-#if DPDK_VIRTIO_WRITER_STATS_COLLECT == 1
+#ifdef RTE_PORT_STATS_COLLECT
 
 #define DPDK_VIRTIO_WRITER_STATS_PKTS_IN_ADD(port, val) \
         port->stats.n_pkts_in += val
@@ -754,7 +754,7 @@ dpdk_virtio_dev_to_vm_tx_burst(struct dpdk_virtio_writer *p,
         while (total_copied < pkt_len) {
             /* Copy mbuf data to buffer */
             rte_memcpy((void *)(uintptr_t)(buff_addr + vb_offset),
-                rte_pktmbuf_mtod(buff, const void *) + offset,
+                rte_pktmbuf_mtod_offset(buff, const void *, offset),
                 len_to_cpy);
 
             offset += len_to_cpy;

diff --git a/include/vr_dpdk.h b/include/vr_dpdk.h
@@ -18,6 +18,7 @@
 #define _VR_DPDK_H_
 
 #include "vr_os.h"
+#include "vr_dpdk_compat.h"
 #include "vr_interface.h"
 #include "vr_packet.h"
 #include "vr_fragment.h"
@@ -477,14 +478,6 @@ struct vr_dpdk_global {
 
 extern struct vr_dpdk_global vr_dpdk;
 
-/**
- *  Enable sent/received/dropped packets statistics
- */
-#define DPDK_KNIDEV_WRITER_STATS_COLLECT    1
-#define DPDK_KNIDEV_READER_STATS_COLLECT    1
-#define DPDK_VIRTIO_WRITER_STATS_COLLECT    1
-#define DPDK_VIRTIO_READER_STATS_COLLECT    1
-
 /*
  * rte_mbuf <=> vr_packet conversion
  *
@@ -540,7 +533,9 @@ vr_dpdk_mbuf_reset(struct vr_packet *pkt)
 /*
  * dpdk_vrouter.c
  */
-/* pktmbuf constructor with vr_packet support */
+/* A packet mbuf pool constructor with vr_packet support */
+void vr_dpdk_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
+/* The packet mbuf constructor with vr_packet support */
 void vr_dpdk_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg, void *_m, unsigned i);
 /* Check if the stop flag is set */
 int vr_dpdk_is_stop_flag_set(void);

diff --git a/include/vr_dpdk_compat.h b/include/vr_dpdk_compat.h
@@ -0,0 +1,53 @@
+/*
+ * vr_dpdk_compat.h - DPDK compatibility definitions
+ *
+ * Copyright (c) 2015 Semihalf.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+
+#include <rte_version.h>
+
+#ifndef __VRDPDKCOMPAT_H__
+#define __VRDPDKCOMPAT_H__
+
+/*
+ * DPDK 2.1
+ */
+#if (RTE_VERSION >= RTE_VERSION_NUM(2, 1, 0, 0))
+
+/*
+ * DPDK 2.0
+ */
+#elif (RTE_VERSION >= RTE_VERSION_NUM(2, 0, 0, 0))
+
+/* Enable port statistics. */
+#define RTE_PORT_STATS_COLLECT
+
+/**
+ * A macro that points to an offset into the data in the mbuf.
+ *
+ * The returned pointer is cast to type t. Before using this
+ * function, the user must ensure that the first segment is large
+ * enough to accommodate its data.
+ *
+ * @param m
+ *   The packet mbuf.
+ * @param o
+ *   The offset into the mbuf data.
+ * @param t
+ *   The type to cast the result into.
+ */
+#define rte_pktmbuf_mtod_offset(m, t, o) \
+        ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
+
+#endif /* RTE_VERSION */
+
+#endif /* __VRDPDKCOMPAT_H__ */