From mboxrd@z Thu Jan 1 00:00:00 1970 From: Michael Chan Subject: [PATCH net 4/5] bnxt_en: Fix implementation of tx push operation. Date: Wed, 10 Feb 2016 17:33:49 -0500 Message-ID: <1455143630-30140-5-git-send-email-mchan@broadcom.com> References: <1455143630-30140-1-git-send-email-mchan@broadcom.com> Cc: netdev@vger.kernel.org To: davem@davemloft.net Return-path: Received: from 5520-maca-inet1-outside.broadcom.com ([216.31.211.11]:59459 "EHLO mail-irv-18.broadcom.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750828AbcBJWdv (ORCPT ); Wed, 10 Feb 2016 17:33:51 -0500 In-Reply-To: <1455143630-30140-1-git-send-email-mchan@broadcom.com> Sender: netdev-owner@vger.kernel.org List-ID: tx push is supported for small packets to reduce DMA latency. The following bugs are fixed in this patch: 1. Fix the definition of the push BD which is different from the DMA BD. 2. The push buffer has to be zero padded to the next 64-bit word boundary or tx checksum won't be correct. 3. Increase the tx push packet threshold to 164 bytes (192 bytes with the BD) so that small tunneled packets are within the threshold. Signed-off-by: Michael Chan --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 42 ++++++++++++++++++------------- drivers/net/ethernet/broadcom/bnxt/bnxt.h | 11 ++++++-- 2 files changed, 34 insertions(+), 19 deletions(-) diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index ddcf7ef..82f4e6d 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -69,7 +69,7 @@ MODULE_VERSION(DRV_MODULE_VERSION); #define BNXT_RX_DMA_OFFSET NET_SKB_PAD #define BNXT_RX_COPY_THRESH 256 -#define BNXT_TX_PUSH_THRESH 92 +#define BNXT_TX_PUSH_THRESH 164 enum board_idx { BCM57301, @@ -223,11 +223,12 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev) } if (free_size == bp->tx_ring_size && length <= bp->tx_push_thresh) { - struct tx_push_bd *push = txr->tx_push; - struct tx_bd *tx_push = &push->txbd1; - struct tx_bd_ext *tx_push1 = &push->txbd2; - void *pdata = tx_push1 + 1; - int j; + struct tx_push_buffer *tx_push_buf = txr->tx_push; + struct tx_push_bd *tx_push = &tx_push_buf->push_bd; + struct tx_bd_ext *tx_push1 = &tx_push->txbd2; + void *pdata = tx_push_buf->data; + u64 *end; + int j, push_len; /* Set COAL_NOW to be ready quickly for the next push */ tx_push->tx_bd_len_flags_type = @@ -247,6 +248,9 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev) tx_push1->tx_bd_cfa_meta = cpu_to_le32(vlan_tag_flags); tx_push1->tx_bd_cfa_action = cpu_to_le32(cfa_action); + end = PTR_ALIGN(pdata + length + 1, 8) - 1; + *end = 0; + skb_copy_from_linear_data(skb, pdata, len); pdata += len; for (j = 0; j < last_frag; j++) { @@ -261,22 +265,29 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev) pdata += skb_frag_size(frag); } - memcpy(txbd, tx_push, sizeof(*txbd)); + txbd->tx_bd_len_flags_type = tx_push->tx_bd_len_flags_type; + txbd->tx_bd_haddr = txr->data_mapping; prod = NEXT_TX(prod); txbd = &txr->tx_desc_ring[TX_RING(prod)][TX_IDX(prod)]; memcpy(txbd, tx_push1, sizeof(*txbd)); prod = NEXT_TX(prod); - push->doorbell = + tx_push->doorbell = cpu_to_le32(DB_KEY_TX_PUSH | DB_LONG_TX_PUSH | prod); txr->tx_prod = prod; netdev_tx_sent_queue(txq, skb->len); - __iowrite64_copy(txr->tx_doorbell, push, - (length + sizeof(*push) + 8) / 8); + push_len = (length + sizeof(*tx_push) + 7) / 8; + if (push_len > 16) { + __iowrite64_copy(txr->tx_doorbell, tx_push_buf, 16); + __iowrite64_copy(txr->tx_doorbell + 4, tx_push_buf + 1, + push_len - 16); + } else { + __iowrite64_copy(txr->tx_doorbell, tx_push_buf, + push_len); + } tx_buf->is_push = 1; - goto tx_done; } @@ -1753,7 +1764,7 @@ static int bnxt_alloc_tx_rings(struct bnxt *bp) push_size = L1_CACHE_ALIGN(sizeof(struct tx_push_bd) + bp->tx_push_thresh); - if (push_size > 128) { + if (push_size > 256) { push_size = 0; bp->tx_push_thresh = 0; } @@ -1772,7 +1783,6 @@ static int bnxt_alloc_tx_rings(struct bnxt *bp) return rc; if (bp->tx_push_size) { - struct tx_bd *txbd; dma_addr_t mapping; /* One pre-allocated DMA buffer to backup @@ -1786,13 +1796,11 @@ static int bnxt_alloc_tx_rings(struct bnxt *bp) if (!txr->tx_push) return -ENOMEM; - txbd = &txr->tx_push->txbd1; - mapping = txr->tx_push_mapping + sizeof(struct tx_push_bd); - txbd->tx_bd_haddr = cpu_to_le64(mapping); + txr->data_mapping = cpu_to_le64(mapping); - memset(txbd + 1, 0, sizeof(struct tx_bd_ext)); + memset(txr->tx_push, 0, sizeof(struct tx_push_bd)); } ring->queue_id = bp->q_info[j].queue_id; if (i % bp->tx_nr_rings_per_tc == (bp->tx_nr_rings_per_tc - 1)) diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h index 8af3ca8..9b4866c 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h @@ -523,10 +523,16 @@ struct bnxt_ring_struct { struct tx_push_bd { __le32 doorbell; - struct tx_bd txbd1; + __le32 tx_bd_len_flags_type; + u32 tx_bd_opaque; struct tx_bd_ext txbd2; }; +struct tx_push_buffer { + struct tx_push_bd push_bd; + u32 data[25]; +}; + struct bnxt_tx_ring_info { struct bnxt_napi *bnapi; u16 tx_prod; @@ -538,8 +544,9 @@ struct bnxt_tx_ring_info { dma_addr_t tx_desc_mapping[MAX_TX_PAGES]; - struct tx_push_bd *tx_push; + struct tx_push_buffer *tx_push; dma_addr_t tx_push_mapping; + __le64 data_mapping; #define BNXT_DEV_STATE_CLOSING 0x1 u32 dev_state; -- 1.8.3.1