[PATCH net-next 11/16] net/mlx5e: Xmit flow break down
From: Saeed Mahameed <hidden>
Date: 2017-04-13 03:37:01
Also in:
linux-rdma
Subsystem:
mellanox ethernet driver (mlx5e), mellanox mlx5 core vpi driver, networking drivers, the rest · Maintainers:
Saeed Mahameed, Tariq Toukan, Mark Bloch, Leon Romanovsky, Andrew Lunn, "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds
Break current mlx5e xmit flow into smaller blocks (helper functions) in order to reuse them for IPoIB SKB transmission. Signed-off-by: Saeed Mahameed <saeedm-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> Reviewed-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> --- drivers/net/ethernet/mellanox/mlx5/core/en.h | 2 +- drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 7 +- drivers/net/ethernet/mellanox/mlx5/core/en_tx.c | 199 +++++++++++++--------- 3 files changed, 119 insertions(+), 89 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 23b92ec54e12..25185f8c3562 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h@@ -304,6 +304,7 @@ struct mlx5e_cq { } ____cacheline_aligned_in_smp; struct mlx5e_tx_wqe_info { + struct sk_buff *skb; u32 num_bytes; u8 num_wqebbs; u8 num_dma;
@@ -345,7 +346,6 @@ struct mlx5e_txqsq { /* write@xmit, read@completion */ struct { - struct sk_buff **skb; struct mlx5e_sq_dma *dma_fifo; struct mlx5e_tx_wqe_info *wqe_info; } db;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index eb657987e9b5..2201b7ea05f4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c@@ -1042,7 +1042,6 @@ static void mlx5e_free_txqsq_db(struct mlx5e_txqsq *sq) { kfree(sq->db.wqe_info); kfree(sq->db.dma_fifo); - kfree(sq->db.skb); } static int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa)
@@ -1050,13 +1049,11 @@ static int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa) int wq_sz = mlx5_wq_cyc_get_size(&sq->wq); int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS; - sq->db.skb = kzalloc_node(wq_sz * sizeof(*sq->db.skb), - GFP_KERNEL, numa); sq->db.dma_fifo = kzalloc_node(df_sz * sizeof(*sq->db.dma_fifo), GFP_KERNEL, numa); sq->db.wqe_info = kzalloc_node(wq_sz * sizeof(*sq->db.wqe_info), GFP_KERNEL, numa); - if (!sq->db.skb || !sq->db.dma_fifo || !sq->db.wqe_info) { + if (!sq->db.dma_fifo || !sq->db.wqe_info) { mlx5e_free_txqsq_db(sq); return -ENOMEM; }
@@ -1295,7 +1292,7 @@ static void mlx5e_deactivate_txqsq(struct mlx5e_txqsq *sq) if (mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, 1)) { struct mlx5e_tx_wqe *nop; - sq->db.skb[(sq->pc & sq->wq.sz_m1)] = NULL; + sq->db.wqe_info[(sq->pc & sq->wq.sz_m1)].skb = NULL; nop = mlx5e_post_nop(&sq->wq, sq->sqn, &sq->pc); mlx5e_notify_hw(&sq->wq, sq->pc, sq->uar_map, &nop->ctrl); }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 5bbc313e70c5..ba664a1126cf 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c@@ -177,30 +177,9 @@ static inline void mlx5e_insert_vlan(void *start, struct sk_buff *skb, u16 ihs, mlx5e_tx_skb_pull_inline(skb_data, skb_len, cpy2_sz); } -static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb) +static inline void +mlx5e_txwqe_build_eseg_csum(struct mlx5e_txqsq *sq, struct sk_buff *skb, struct mlx5_wqe_eth_seg *eseg) { - struct mlx5_wq_cyc *wq = &sq->wq; - - u16 pi = sq->pc & wq->sz_m1; - struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(wq, pi); - struct mlx5e_tx_wqe_info *wi = &sq->db.wqe_info[pi]; - - struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl; - struct mlx5_wqe_eth_seg *eseg = &wqe->eth; - struct mlx5_wqe_data_seg *dseg; - - unsigned char *skb_data = skb->data; - unsigned int skb_len = skb->len; - u8 opcode = MLX5_OPCODE_SEND; - dma_addr_t dma_addr = 0; - unsigned int num_bytes; - u16 headlen; - u16 ds_cnt; - u16 ihs; - int i; - - memset(wqe, 0, sizeof(*wqe)); - if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) { eseg->cs_flags = MLX5_ETH_WQE_L3_CSUM; if (skb->encapsulation) {
@@ -212,66 +191,51 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb) } } else sq->stats.csum_none++; +} - if (skb_is_gso(skb)) { - eseg->mss = cpu_to_be16(skb_shinfo(skb)->gso_size); - opcode = MLX5_OPCODE_LSO; +static inline u16 +mlx5e_txwqe_build_eseg_gso(struct mlx5e_txqsq *sq, struct sk_buff *skb, + struct mlx5_wqe_eth_seg *eseg, unsigned int *num_bytes) +{ + u16 ihs; - if (skb->encapsulation) { - ihs = skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb); - sq->stats.tso_inner_packets++; - sq->stats.tso_inner_bytes += skb->len - ihs; - } else { - ihs = skb_transport_offset(skb) + tcp_hdrlen(skb); - sq->stats.tso_packets++; - sq->stats.tso_bytes += skb->len - ihs; - } + eseg->mss = cpu_to_be16(skb_shinfo(skb)->gso_size); - sq->stats.packets += skb_shinfo(skb)->gso_segs; - num_bytes = skb->len + (skb_shinfo(skb)->gso_segs - 1) * ihs; + if (skb->encapsulation) { + ihs = skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb); + sq->stats.tso_inner_packets++; + sq->stats.tso_inner_bytes += skb->len - ihs; } else { - ihs = mlx5e_calc_min_inline(sq->min_inline_mode, skb); - sq->stats.packets++; - num_bytes = max_t(unsigned int, skb->len, ETH_ZLEN); + ihs = skb_transport_offset(skb) + tcp_hdrlen(skb); + sq->stats.tso_packets++; + sq->stats.tso_bytes += skb->len - ihs; } - sq->stats.bytes += num_bytes; - wi->num_bytes = num_bytes; - - ds_cnt = sizeof(*wqe) / MLX5_SEND_WQE_DS; - if (ihs) { - if (skb_vlan_tag_present(skb)) { - mlx5e_insert_vlan(eseg->inline_hdr.start, skb, ihs, &skb_data, &skb_len); - ihs += VLAN_HLEN; - } else { - memcpy(eseg->inline_hdr.start, skb_data, ihs); - mlx5e_tx_skb_pull_inline(&skb_data, &skb_len, ihs); - } - eseg->inline_hdr.sz = cpu_to_be16(ihs); - ds_cnt += DIV_ROUND_UP(ihs - sizeof(eseg->inline_hdr.start), MLX5_SEND_WQE_DS); - } else if (skb_vlan_tag_present(skb)) { - eseg->insert.type = cpu_to_be16(MLX5_ETH_WQE_INSERT_VLAN); - eseg->insert.vlan_tci = cpu_to_be16(skb_vlan_tag_get(skb)); - } - - dseg = (struct mlx5_wqe_data_seg *)cseg + ds_cnt; + *num_bytes = skb->len + (skb_shinfo(skb)->gso_segs - 1) * ihs; + return ihs; +} - wi->num_dma = 0; +static inline int +mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, + unsigned char *skb_data, u16 headlen, + struct mlx5_wqe_data_seg *dseg) +{ + dma_addr_t dma_addr = 0; + u8 num_dma = 0; + int i; - headlen = skb_len - skb->data_len; if (headlen) { dma_addr = dma_map_single(sq->pdev, skb_data, headlen, DMA_TO_DEVICE); if (unlikely(dma_mapping_error(sq->pdev, dma_addr))) - goto dma_unmap_wqe_err; + return -ENOMEM; dseg->addr = cpu_to_be64(dma_addr); dseg->lkey = sq->mkey_be; dseg->byte_count = cpu_to_be32(headlen); mlx5e_dma_push(sq, dma_addr, headlen, MLX5E_DMA_MAP_SINGLE); - wi->num_dma++; - + num_dma++; dseg++; }
@@ -280,51 +244,120 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb) int fsz = skb_frag_size(frag); dma_addr = skb_frag_dma_map(sq->pdev, frag, 0, fsz, - DMA_TO_DEVICE); + DMA_TO_DEVICE); if (unlikely(dma_mapping_error(sq->pdev, dma_addr))) - goto dma_unmap_wqe_err; + return -ENOMEM; dseg->addr = cpu_to_be64(dma_addr); dseg->lkey = sq->mkey_be; dseg->byte_count = cpu_to_be32(fsz); mlx5e_dma_push(sq, dma_addr, fsz, MLX5E_DMA_MAP_PAGE); - wi->num_dma++; - + num_dma++; dseg++; } - ds_cnt += wi->num_dma; - - cseg->opmod_idx_opcode = cpu_to_be32((sq->pc << 8) | opcode); - cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt); + return num_dma; +} - sq->db.skb[pi] = skb; +static inline void +mlx5e_txwqe_complete(struct mlx5e_txqsq *sq, struct sk_buff *skb, + u8 opcode, u16 ds_cnt, u32 num_bytes, u8 num_dma, + struct mlx5e_tx_wqe_info *wi, struct mlx5_wqe_ctrl_seg *cseg) +{ + struct mlx5_wq_cyc *wq = &sq->wq; + u16 pi; + wi->num_bytes = num_bytes; + wi->num_dma = num_dma; wi->num_wqebbs = DIV_ROUND_UP(ds_cnt, MLX5_SEND_WQEBB_NUM_DS); - sq->pc += wi->num_wqebbs; + wi->skb = skb; - netdev_tx_sent_queue(sq->txq, wi->num_bytes); + cseg->opmod_idx_opcode = cpu_to_be32((sq->pc << 8) | opcode); + cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt); + + netdev_tx_sent_queue(sq->txq, num_bytes); if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; - if (unlikely(!mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, - MLX5E_SQ_STOP_ROOM))) { + sq->pc += wi->num_wqebbs; + if (unlikely(!mlx5e_wqc_has_room_for(wq, sq->cc, sq->pc, MLX5E_SQ_STOP_ROOM))) { netif_tx_stop_queue(sq->txq); sq->stats.stopped++; } - sq->stats.xmit_more += skb->xmit_more; if (!skb->xmit_more || netif_xmit_stopped(sq->txq)) mlx5e_notify_hw(wq, sq->pc, sq->uar_map, cseg); /* fill sq edge with nops to avoid wqe wrap around */ while ((pi = (sq->pc & wq->sz_m1)) > sq->edge) { - sq->db.skb[pi] = NULL; - mlx5e_post_nop(&sq->wq, sq->sqn, &sq->pc); + sq->db.wqe_info[pi].skb = NULL; + mlx5e_post_nop(wq, sq->sqn, &sq->pc); sq->stats.nop++; } +} + +static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb) +{ + struct mlx5_wq_cyc *wq = &sq->wq; + + u16 pi = sq->pc & wq->sz_m1; + struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(wq, pi); + struct mlx5e_tx_wqe_info *wi = &sq->db.wqe_info[pi]; + + struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl; + struct mlx5_wqe_eth_seg *eseg = &wqe->eth; + + unsigned char *skb_data = skb->data; + unsigned int skb_len = skb->len; + u8 opcode = MLX5_OPCODE_SEND; + unsigned int num_bytes; + int num_dma; + u16 headlen; + u16 ds_cnt; + u16 ihs; + + memset(wqe, 0, sizeof(*wqe)); + + mlx5e_txwqe_build_eseg_csum(sq, skb, eseg); + + if (skb_is_gso(skb)) { + opcode = MLX5_OPCODE_LSO; + ihs = mlx5e_txwqe_build_eseg_gso(sq, skb, eseg, &num_bytes); + sq->stats.packets += skb_shinfo(skb)->gso_segs; + } else { + ihs = mlx5e_calc_min_inline(sq->min_inline_mode, skb); + num_bytes = max_t(unsigned int, skb->len, ETH_ZLEN); + sq->stats.packets++; + } + sq->stats.bytes += num_bytes; + sq->stats.xmit_more += skb->xmit_more; + + ds_cnt = sizeof(*wqe) / MLX5_SEND_WQE_DS; + if (ihs) { + if (skb_vlan_tag_present(skb)) { + mlx5e_insert_vlan(eseg->inline_hdr.start, skb, ihs, &skb_data, &skb_len); + ihs += VLAN_HLEN; + } else { + memcpy(eseg->inline_hdr.start, skb_data, ihs); + mlx5e_tx_skb_pull_inline(&skb_data, &skb_len, ihs); + } + eseg->inline_hdr.sz = cpu_to_be16(ihs); + ds_cnt += DIV_ROUND_UP(ihs - sizeof(eseg->inline_hdr.start), MLX5_SEND_WQE_DS); + } else if (skb_vlan_tag_present(skb)) { + eseg->insert.type = cpu_to_be16(MLX5_ETH_WQE_INSERT_VLAN); + eseg->insert.vlan_tci = cpu_to_be16(skb_vlan_tag_get(skb)); + } + + headlen = skb_len - skb->data_len; + num_dma = mlx5e_txwqe_build_dsegs(sq, skb, skb_data, headlen, + (struct mlx5_wqe_data_seg *)cseg + ds_cnt); + if (unlikely(num_dma < 0)) + goto dma_unmap_wqe_err; + + mlx5e_txwqe_complete(sq, skb, opcode, ds_cnt + num_dma, + num_bytes, num_dma, wi, cseg); return NETDEV_TX_OK;
@@ -392,8 +425,8 @@ bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget) last_wqe = (sqcc == wqe_counter); ci = sqcc & sq->wq.sz_m1; - skb = sq->db.skb[ci]; wi = &sq->db.wqe_info[ci]; + skb = wi->skb; if (unlikely(!skb)) { /* nop */ sqcc++;
@@ -451,8 +484,8 @@ void mlx5e_free_txqsq_descs(struct mlx5e_txqsq *sq) while (sq->cc != sq->pc) { ci = sq->cc & sq->wq.sz_m1; - skb = sq->db.skb[ci]; wi = &sq->db.wqe_info[ci]; + skb = wi->skb; if (!skb) { /* nop */ sq->cc++;
--
2.11.0
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html