net: atlantic: Implement xdp data plane
authorTaehee Yoo <ap420073@gmail.com>
Sun, 17 Apr 2022 10:12:46 +0000 (10:12 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 20 Apr 2022 09:42:57 +0000 (10:42 +0100)
It supports XDP_PASS, XDP_DROP and multi buffer.

The new function aq_nic_xmit_xdpf() is used to send packet with
xdp_frame and internally it calls aq_nic_map_xdp().

AQC chip supports 32 multi-queues and 8 vectors(irq).
there are two option
1. under 8 cores and 4 tx queues per core.
2. under 4 cores and 8 tx queues per core.

Like ixgbe, these tx queues can be used only for XDP_TX, XDP_REDIRECT
queue. If so, no tx_lock is needed.
But this patchset doesn't use this strategy because getting hardware tx
queue index cost is too high.
So, tx_lock is used in the aq_nic_xmit_xdpf().

single-core, single queue, 80% cpu utilization.

  30.75%  bpf_prog_xxx_xdp_prog_tx  [k] bpf_prog_xxx_xdp_prog_tx
  10.35%  [kernel]                  [k] aq_hw_read_reg <---------- here
   4.38%  [kernel]                  [k] get_page_from_freelist

single-core, 8 queues, 100% cpu utilization, half PPS.

  45.56%  [kernel]                  [k] aq_hw_read_reg <---------- here
  17.58%  bpf_prog_xxx_xdp_prog_tx  [k] bpf_prog_xxx_xdp_prog_tx
   4.72%  [kernel]                  [k] hw_atl_b0_hw_ring_rx_receive

The new function __aq_ring_xdp_clean() is a xdp rx handler and this is
called only when XDP is attached.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/aquantia/atlantic/aq_ethtool.c
drivers/net/ethernet/aquantia/atlantic/aq_nic.c
drivers/net/ethernet/aquantia/atlantic/aq_nic.h
drivers/net/ethernet/aquantia/atlantic/aq_ring.c
drivers/net/ethernet/aquantia/atlantic/aq_ring.h
drivers/net/ethernet/aquantia/atlantic/aq_vec.c

index a418238..1daecd4 100644 (file)
@@ -97,6 +97,15 @@ static const char * const aq_ethtool_queue_rx_stat_names[] = {
        "%sQueue[%d] AllocFails",
        "%sQueue[%d] SkbAllocFails",
        "%sQueue[%d] Polls",
+       "%sQueue[%d] PageFlips",
+       "%sQueue[%d] PageReuses",
+       "%sQueue[%d] PageFrees",
+       "%sQueue[%d] XdpAbort",
+       "%sQueue[%d] XdpDrop",
+       "%sQueue[%d] XdpPass",
+       "%sQueue[%d] XdpTx",
+       "%sQueue[%d] XdpInvalid",
+       "%sQueue[%d] XdpRedirect",
 };
 
 static const char * const aq_ethtool_queue_tx_stat_names[] = {
index 24d715c..e11cc29 100644 (file)
@@ -569,6 +569,103 @@ err_exit:
        return err;
 }
 
+static unsigned int aq_nic_map_xdp(struct aq_nic_s *self,
+                                  struct xdp_frame *xdpf,
+                                  struct aq_ring_s *ring)
+{
+       struct device *dev = aq_nic_get_dev(self);
+       struct aq_ring_buff_s *first = NULL;
+       unsigned int dx = ring->sw_tail;
+       struct aq_ring_buff_s *dx_buff;
+       struct skb_shared_info *sinfo;
+       unsigned int frag_count = 0U;
+       unsigned int nr_frags = 0U;
+       unsigned int ret = 0U;
+       u16 total_len;
+
+       dx_buff = &ring->buff_ring[dx];
+       dx_buff->flags = 0U;
+
+       sinfo = xdp_get_shared_info_from_frame(xdpf);
+       total_len = xdpf->len;
+       dx_buff->len = total_len;
+       if (xdp_frame_has_frags(xdpf)) {
+               nr_frags = sinfo->nr_frags;
+               total_len += sinfo->xdp_frags_size;
+       }
+       dx_buff->pa = dma_map_single(dev, xdpf->data, dx_buff->len,
+                                    DMA_TO_DEVICE);
+
+       if (unlikely(dma_mapping_error(dev, dx_buff->pa)))
+               goto exit;
+
+       first = dx_buff;
+       dx_buff->len_pkt = total_len;
+       dx_buff->is_sop = 1U;
+       dx_buff->is_mapped = 1U;
+       ++ret;
+
+       for (; nr_frags--; ++frag_count) {
+               skb_frag_t *frag = &sinfo->frags[frag_count];
+               unsigned int frag_len = skb_frag_size(frag);
+               unsigned int buff_offset = 0U;
+               unsigned int buff_size = 0U;
+               dma_addr_t frag_pa;
+
+               while (frag_len) {
+                       if (frag_len > AQ_CFG_TX_FRAME_MAX)
+                               buff_size = AQ_CFG_TX_FRAME_MAX;
+                       else
+                               buff_size = frag_len;
+
+                       frag_pa = skb_frag_dma_map(dev, frag, buff_offset,
+                                                  buff_size, DMA_TO_DEVICE);
+
+                       if (unlikely(dma_mapping_error(dev, frag_pa)))
+                               goto mapping_error;
+
+                       dx = aq_ring_next_dx(ring, dx);
+                       dx_buff = &ring->buff_ring[dx];
+
+                       dx_buff->flags = 0U;
+                       dx_buff->len = buff_size;
+                       dx_buff->pa = frag_pa;
+                       dx_buff->is_mapped = 1U;
+                       dx_buff->eop_index = 0xffffU;
+
+                       frag_len -= buff_size;
+                       buff_offset += buff_size;
+
+                       ++ret;
+               }
+       }
+
+       first->eop_index = dx;
+       dx_buff->is_eop = 1U;
+       dx_buff->skb = NULL;
+       dx_buff->xdpf = xdpf;
+       goto exit;
+
+mapping_error:
+       for (dx = ring->sw_tail;
+            ret > 0;
+            --ret, dx = aq_ring_next_dx(ring, dx)) {
+               dx_buff = &ring->buff_ring[dx];
+
+               if (!dx_buff->pa)
+                       continue;
+               if (unlikely(dx_buff->is_sop))
+                       dma_unmap_single(dev, dx_buff->pa, dx_buff->len,
+                                        DMA_TO_DEVICE);
+               else
+                       dma_unmap_page(dev, dx_buff->pa, dx_buff->len,
+                                      DMA_TO_DEVICE);
+       }
+
+exit:
+       return ret;
+}
+
 unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
                            struct aq_ring_s *ring)
 {
@@ -697,6 +794,7 @@ unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
        first->eop_index = dx;
        dx_buff->is_eop = 1U;
        dx_buff->skb = skb;
+       dx_buff->xdpf = NULL;
        goto exit;
 
 mapping_error:
@@ -725,6 +823,44 @@ exit:
        return ret;
 }
 
+int aq_nic_xmit_xdpf(struct aq_nic_s *aq_nic, struct aq_ring_s *tx_ring,
+                    struct xdp_frame *xdpf)
+{
+       u16 queue_index = AQ_NIC_RING2QMAP(aq_nic, tx_ring->idx);
+       struct net_device *ndev = aq_nic_get_ndev(aq_nic);
+       struct skb_shared_info *sinfo;
+       int cpu = smp_processor_id();
+       int err = NETDEV_TX_BUSY;
+       struct netdev_queue *nq;
+       unsigned int frags = 1;
+
+       if (xdp_frame_has_frags(xdpf)) {
+               sinfo = xdp_get_shared_info_from_frame(xdpf);
+               frags += sinfo->nr_frags;
+       }
+
+       if (frags > AQ_CFG_SKB_FRAGS_MAX)
+               return err;
+
+       nq = netdev_get_tx_queue(ndev, tx_ring->idx);
+       __netif_tx_lock(nq, cpu);
+
+       aq_ring_update_queue_state(tx_ring);
+
+       /* Above status update may stop the queue. Check this. */
+       if (__netif_subqueue_stopped(aq_nic_get_ndev(aq_nic), queue_index))
+               goto out;
+
+       frags = aq_nic_map_xdp(aq_nic, xdpf, tx_ring);
+       if (likely(frags))
+               err = aq_nic->aq_hw_ops->hw_ring_tx_xmit(aq_nic->aq_hw, tx_ring,
+                                                        frags);
+out:
+       __netif_tx_unlock(nq);
+
+       return err;
+}
+
 int aq_nic_xmit(struct aq_nic_s *self, struct sk_buff *skb)
 {
        struct aq_nic_cfg_s *cfg = aq_nic_get_cfg(self);
index 47123ba..935ba88 100644 (file)
@@ -180,6 +180,8 @@ void aq_nic_ndev_free(struct aq_nic_s *self);
 int aq_nic_start(struct aq_nic_s *self);
 unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
                            struct aq_ring_s *ring);
+int aq_nic_xmit_xdpf(struct aq_nic_s *aq_nic, struct aq_ring_s *tx_ring,
+                    struct xdp_frame *xdpf);
 int aq_nic_xmit(struct aq_nic_s *self, struct sk_buff *skb);
 int aq_nic_get_regs(struct aq_nic_s *self, struct ethtool_regs *regs, void *p);
 int aq_nic_get_regs_count(struct aq_nic_s *self);
index 31a14a0..9bb9b76 100644 (file)
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 
+static void aq_get_rxpages_xdp(struct aq_ring_buff_s *buff,
+                              struct xdp_buff *xdp)
+{
+       struct skb_shared_info *sinfo;
+       int i;
+
+       if (xdp_buff_has_frags(xdp)) {
+               sinfo = xdp_get_shared_info_from_buff(xdp);
+
+               for (i = 0; i < sinfo->nr_frags; i++) {
+                       skb_frag_t *frag = &sinfo->frags[i];
+
+                       page_ref_inc(skb_frag_page(frag));
+               }
+       }
+       page_ref_inc(buff->rxdata.page);
+}
+
 static inline void aq_free_rxpage(struct aq_rxpage *rxpage, struct device *dev)
 {
        unsigned int len = PAGE_SIZE << rxpage->order;
@@ -319,15 +337,26 @@ bool aq_ring_tx_clean(struct aq_ring_s *self)
                        }
                }
 
-               if (unlikely(buff->is_eop && buff->skb)) {
+               if (likely(!buff->is_eop))
+                       goto out;
+
+               if (buff->skb) {
                        u64_stats_update_begin(&self->stats.tx.syncp);
                        ++self->stats.tx.packets;
                        self->stats.tx.bytes += buff->skb->len;
                        u64_stats_update_end(&self->stats.tx.syncp);
-
                        dev_kfree_skb_any(buff->skb);
-                       buff->skb = NULL;
+               } else if (buff->xdpf) {
+                       u64_stats_update_begin(&self->stats.tx.syncp);
+                       ++self->stats.tx.packets;
+                       self->stats.tx.bytes += xdp_get_frame_len(buff->xdpf);
+                       u64_stats_update_end(&self->stats.tx.syncp);
+                       xdp_return_frame_rx_napi(buff->xdpf);
                }
+
+out:
+               buff->skb = NULL;
+               buff->xdpf = NULL;
                buff->pa = 0U;
                buff->eop_index = 0xffffU;
                self->sw_head = aq_ring_next_dx(self, self->sw_head);
@@ -360,11 +389,136 @@ static void aq_rx_checksum(struct aq_ring_s *self,
                __skb_incr_checksum_unnecessary(skb);
 }
 
-#define AQ_SKB_ALIGN SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
-int aq_ring_rx_clean(struct aq_ring_s *self,
-                    struct napi_struct *napi,
-                    int *work_done,
-                    int budget)
+static struct sk_buff *aq_xdp_run_prog(struct aq_nic_s *aq_nic,
+                                      struct xdp_buff *xdp,
+                                      struct aq_ring_s *rx_ring,
+                                      struct aq_ring_buff_s *buff)
+{
+       int result = NETDEV_TX_BUSY;
+       struct aq_ring_s *tx_ring;
+       struct xdp_frame *xdpf;
+       struct bpf_prog *prog;
+       u32 act = XDP_ABORTED;
+       struct sk_buff *skb;
+
+       u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+       ++rx_ring->stats.rx.packets;
+       rx_ring->stats.rx.bytes += xdp_get_buff_len(xdp);
+       u64_stats_update_end(&rx_ring->stats.rx.syncp);
+
+       prog = READ_ONCE(rx_ring->xdp_prog);
+       if (!prog)
+               goto pass;
+
+       prefetchw(xdp->data_hard_start); /* xdp_frame write */
+
+       /* single buffer XDP program, but packet is multi buffer, aborted */
+       if (xdp_buff_has_frags(xdp) && !prog->aux->xdp_has_frags)
+               goto out_aborted;
+
+       act = bpf_prog_run_xdp(prog, xdp);
+       switch (act) {
+       case XDP_PASS:
+pass:
+               xdpf = xdp_convert_buff_to_frame(xdp);
+               if (unlikely(!xdpf))
+                       goto out_aborted;
+               skb = xdp_build_skb_from_frame(xdpf, aq_nic->ndev);
+               if (!skb)
+                       goto out_aborted;
+               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+               ++rx_ring->stats.rx.xdp_pass;
+               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+               aq_get_rxpages_xdp(buff, xdp);
+               return skb;
+       case XDP_TX:
+               xdpf = xdp_convert_buff_to_frame(xdp);
+               if (unlikely(!xdpf))
+                       goto out_aborted;
+               tx_ring = aq_nic->aq_ring_tx[rx_ring->idx];
+               result = aq_nic_xmit_xdpf(aq_nic, tx_ring, xdpf);
+               if (result == NETDEV_TX_BUSY)
+                       goto out_aborted;
+               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+               ++rx_ring->stats.rx.xdp_tx;
+               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+               aq_get_rxpages_xdp(buff, xdp);
+               break;
+       case XDP_REDIRECT:
+               if (xdp_do_redirect(aq_nic->ndev, xdp, prog) < 0)
+                       goto out_aborted;
+               xdp_do_flush();
+               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+               ++rx_ring->stats.rx.xdp_redirect;
+               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+               aq_get_rxpages_xdp(buff, xdp);
+               break;
+       default:
+               fallthrough;
+       case XDP_ABORTED:
+out_aborted:
+               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+               ++rx_ring->stats.rx.xdp_aborted;
+               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+               trace_xdp_exception(aq_nic->ndev, prog, act);
+               bpf_warn_invalid_xdp_action(aq_nic->ndev, prog, act);
+               break;
+       case XDP_DROP:
+               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+               ++rx_ring->stats.rx.xdp_drop;
+               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+               break;
+       }
+
+       return ERR_PTR(-result);
+}
+
+static bool aq_add_rx_fragment(struct device *dev,
+                              struct aq_ring_s *ring,
+                              struct aq_ring_buff_s *buff,
+                              struct xdp_buff *xdp)
+{
+       struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
+       struct aq_ring_buff_s *buff_ = buff;
+
+       memset(sinfo, 0, sizeof(*sinfo));
+       do {
+               skb_frag_t *frag;
+
+               if (unlikely(sinfo->nr_frags >= MAX_SKB_FRAGS))
+                       return true;
+
+               frag = &sinfo->frags[sinfo->nr_frags++];
+               buff_ = &ring->buff_ring[buff_->next];
+               dma_sync_single_range_for_cpu(dev,
+                                             buff_->rxdata.daddr,
+                                             buff_->rxdata.pg_off,
+                                             buff_->len,
+                                             DMA_FROM_DEVICE);
+               skb_frag_off_set(frag, buff_->rxdata.pg_off);
+               skb_frag_size_set(frag, buff_->len);
+               sinfo->xdp_frags_size += buff_->len;
+               __skb_frag_set_page(frag, buff_->rxdata.page);
+
+               buff_->is_cleaned = 1;
+
+               buff->is_ip_cso &= buff_->is_ip_cso;
+               buff->is_udp_cso &= buff_->is_udp_cso;
+               buff->is_tcp_cso &= buff_->is_tcp_cso;
+               buff->is_cso_err |= buff_->is_cso_err;
+
+               if (page_is_pfmemalloc(buff_->rxdata.page))
+                       xdp_buff_set_frag_pfmemalloc(xdp);
+
+       } while (!buff_->is_eop);
+
+       xdp_buff_set_frags_flag(xdp);
+
+       return false;
+}
+
+static int __aq_ring_rx_clean(struct aq_ring_s *self, struct napi_struct *napi,
+                             int *work_done, int budget)
 {
        struct net_device *ndev = aq_nic_get_ndev(self->aq_nic);
        bool is_rsc_completed = true;
@@ -531,6 +685,149 @@ err_exit:
        return err;
 }
 
+static int __aq_ring_xdp_clean(struct aq_ring_s *rx_ring,
+                              struct napi_struct *napi, int *work_done,
+                              int budget)
+{
+       int frame_sz = rx_ring->page_offset + rx_ring->frame_max +
+                      rx_ring->tail_size;
+       struct aq_nic_s *aq_nic = rx_ring->aq_nic;
+       bool is_rsc_completed = true;
+       struct device *dev;
+       int err = 0;
+
+       dev = aq_nic_get_dev(aq_nic);
+       for (; (rx_ring->sw_head != rx_ring->hw_head) && budget;
+               rx_ring->sw_head = aq_ring_next_dx(rx_ring, rx_ring->sw_head),
+               --budget, ++(*work_done)) {
+               struct aq_ring_buff_s *buff = &rx_ring->buff_ring[rx_ring->sw_head];
+               bool is_ptp_ring = aq_ptp_ring(rx_ring->aq_nic, rx_ring);
+               struct aq_ring_buff_s *buff_ = NULL;
+               struct sk_buff *skb = NULL;
+               unsigned int next_ = 0U;
+               struct xdp_buff xdp;
+               void *hard_start;
+
+               if (buff->is_cleaned)
+                       continue;
+
+               if (!buff->is_eop) {
+                       buff_ = buff;
+                       do {
+                               if (buff_->next >= rx_ring->size) {
+                                       err = -EIO;
+                                       goto err_exit;
+                               }
+                               next_ = buff_->next;
+                               buff_ = &rx_ring->buff_ring[next_];
+                               is_rsc_completed =
+                                       aq_ring_dx_in_range(rx_ring->sw_head,
+                                                           next_,
+                                                           rx_ring->hw_head);
+
+                               if (unlikely(!is_rsc_completed))
+                                       break;
+
+                               buff->is_error |= buff_->is_error;
+                               buff->is_cso_err |= buff_->is_cso_err;
+                       } while (!buff_->is_eop);
+
+                       if (!is_rsc_completed) {
+                               err = 0;
+                               goto err_exit;
+                       }
+                       if (buff->is_error ||
+                           (buff->is_lro && buff->is_cso_err)) {
+                               buff_ = buff;
+                               do {
+                                       if (buff_->next >= rx_ring->size) {
+                                               err = -EIO;
+                                               goto err_exit;
+                                       }
+                                       next_ = buff_->next;
+                                       buff_ = &rx_ring->buff_ring[next_];
+
+                                       buff_->is_cleaned = true;
+                               } while (!buff_->is_eop);
+
+                               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+                               ++rx_ring->stats.rx.errors;
+                               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+                               continue;
+                       }
+               }
+
+               if (buff->is_error) {
+                       u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+                       ++rx_ring->stats.rx.errors;
+                       u64_stats_update_end(&rx_ring->stats.rx.syncp);
+                       continue;
+               }
+
+               dma_sync_single_range_for_cpu(dev,
+                                             buff->rxdata.daddr,
+                                             buff->rxdata.pg_off,
+                                             buff->len, DMA_FROM_DEVICE);
+               hard_start = page_address(buff->rxdata.page) +
+                            buff->rxdata.pg_off - rx_ring->page_offset;
+
+               if (is_ptp_ring)
+                       buff->len -=
+                               aq_ptp_extract_ts(rx_ring->aq_nic, skb,
+                                                 aq_buf_vaddr(&buff->rxdata),
+                                                 buff->len);
+
+               xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
+               xdp_prepare_buff(&xdp, hard_start, rx_ring->page_offset,
+                                buff->len, false);
+               if (!buff->is_eop) {
+                       if (aq_add_rx_fragment(dev, rx_ring, buff, &xdp)) {
+                               u64_stats_update_begin(&rx_ring->stats.rx.syncp);
+                               ++rx_ring->stats.rx.packets;
+                               rx_ring->stats.rx.bytes += xdp_get_buff_len(&xdp);
+                               ++rx_ring->stats.rx.xdp_aborted;
+                               u64_stats_update_end(&rx_ring->stats.rx.syncp);
+                               continue;
+                       }
+               }
+
+               skb = aq_xdp_run_prog(aq_nic, &xdp, rx_ring, buff);
+               if (IS_ERR(skb) || !skb)
+                       continue;
+
+               if (buff->is_vlan)
+                       __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
+                                              buff->vlan_rx_tag);
+
+               aq_rx_checksum(rx_ring, buff, skb);
+
+               skb_set_hash(skb, buff->rss_hash,
+                            buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
+                            PKT_HASH_TYPE_NONE);
+               /* Send all PTP traffic to 0 queue */
+               skb_record_rx_queue(skb,
+                                   is_ptp_ring ? 0
+                                               : AQ_NIC_RING2QMAP(rx_ring->aq_nic,
+                                                                  rx_ring->idx));
+
+               napi_gro_receive(napi, skb);
+       }
+
+err_exit:
+       return err;
+}
+
+int aq_ring_rx_clean(struct aq_ring_s *self,
+                    struct napi_struct *napi,
+                    int *work_done,
+                    int budget)
+{
+       if (static_branch_unlikely(&aq_xdp_locking_key))
+               return __aq_ring_xdp_clean(self, napi, work_done, budget);
+       else
+               return __aq_ring_rx_clean(self, napi, work_done, budget);
+}
+
 void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic)
 {
 #if IS_REACHABLE(CONFIG_PTP_1588_CLOCK)
@@ -620,6 +917,15 @@ unsigned int aq_ring_fill_stats_data(struct aq_ring_s *self, u64 *data)
                        data[++count] = self->stats.rx.alloc_fails;
                        data[++count] = self->stats.rx.skb_alloc_fails;
                        data[++count] = self->stats.rx.polls;
+                       data[++count] = self->stats.rx.pg_flips;
+                       data[++count] = self->stats.rx.pg_reuses;
+                       data[++count] = self->stats.rx.pg_losts;
+                       data[++count] = self->stats.rx.xdp_aborted;
+                       data[++count] = self->stats.rx.xdp_drop;
+                       data[++count] = self->stats.rx.xdp_pass;
+                       data[++count] = self->stats.rx.xdp_tx;
+                       data[++count] = self->stats.rx.xdp_invalid;
+                       data[++count] = self->stats.rx.xdp_redirect;
                } while (u64_stats_fetch_retry_irq(&self->stats.rx.syncp, start));
        } else {
                /* This data should mimic aq_ethtool_queue_tx_stat_names structure */
index 2dadfaf..52f40ea 100644 (file)
@@ -106,6 +106,12 @@ struct aq_ring_stats_rx_s {
        u64 pg_losts;
        u64 pg_flips;
        u64 pg_reuses;
+       u64 xdp_aborted;
+       u64 xdp_drop;
+       u64 xdp_pass;
+       u64 xdp_tx;
+       u64 xdp_invalid;
+       u64 xdp_redirect;
 };
 
 struct aq_ring_stats_tx_s {
index 997f351..e839e10 100644 (file)
@@ -155,7 +155,7 @@ int aq_vec_ring_alloc(struct aq_vec_s *self, struct aq_nic_s *aq_nic,
                        goto err_exit;
                }
                if (xdp_rxq_info_reg_mem_model(&self->ring[i][AQ_VEC_RX_ID].xdp_rxq,
-                                              MEM_TYPE_PAGE_ORDER0, NULL) < 0) {
+                                              MEM_TYPE_PAGE_SHARED, NULL) < 0) {
                        xdp_rxq_info_unreg(&self->ring[i][AQ_VEC_RX_ID].xdp_rxq);
                        err = -ENOMEM;
                        goto err_exit;