From 0c729f53b8c33b9e5eadc2d5e673759e3510501e Mon Sep 17 00:00:00 2001 From: Lorenzo Bianconi Date: Tue, 29 Oct 2024 13:17:10 +0100 Subject: [PATCH 2/2] net: airoha: Simplify Tx napi logic Simplify Tx napi logic relying just on the packet index provided by completion queue indicating the completed packet that can be removed from the Tx DMA ring. This is a preliminary patch to add Qdisc offload for airoha_eth driver. Signed-off-by: Lorenzo Bianconi Link: https://patch.msgid.link/20241029-airoha-en7581-tx-napi-work-v1-2-96ad1686b946@kernel.org Signed-off-by: Jakub Kicinski --- drivers/net/ethernet/mediatek/airoha_eth.c | 73 ++++++++++++---------- 1 file changed, 41 insertions(+), 32 deletions(-) --- a/drivers/net/ethernet/mediatek/airoha_eth.c +++ b/drivers/net/ethernet/mediatek/airoha_eth.c @@ -1670,8 +1670,12 @@ static int airoha_qdma_tx_napi_poll(stru irq_queued = FIELD_GET(IRQ_ENTRY_LEN_MASK, status); while (irq_queued > 0 && done < budget) { - u32 qid, last, val = irq_q->q[head]; + u32 qid, val = irq_q->q[head]; + struct airoha_qdma_desc *desc; + struct airoha_queue_entry *e; struct airoha_queue *q; + u32 index, desc_ctrl; + struct sk_buff *skb; if (val == 0xff) break; @@ -1681,9 +1685,7 @@ static int airoha_qdma_tx_napi_poll(stru irq_queued--; done++; - last = FIELD_GET(IRQ_DESC_IDX_MASK, val); qid = FIELD_GET(IRQ_RING_IDX_MASK, val); - if (qid >= ARRAY_SIZE(qdma->q_tx)) continue; @@ -1691,46 +1693,53 @@ static int airoha_qdma_tx_napi_poll(stru if (!q->ndesc) continue; + index = FIELD_GET(IRQ_DESC_IDX_MASK, val); + if (index >= q->ndesc) + continue; + spin_lock_bh(&q->lock); - while (q->queued > 0) { - struct airoha_qdma_desc *desc = &q->desc[q->tail]; - struct airoha_queue_entry *e = &q->entry[q->tail]; - u32 desc_ctrl = le32_to_cpu(desc->ctrl); - struct sk_buff *skb = e->skb; - u16 index = q->tail; - - if (!(desc_ctrl & QDMA_DESC_DONE_MASK) && - !(desc_ctrl & QDMA_DESC_DROP_MASK)) - break; + if (!q->queued) + goto unlock; - q->tail = (q->tail + 1) % q->ndesc; - q->queued--; + desc = &q->desc[index]; + desc_ctrl = le32_to_cpu(desc->ctrl); - dma_unmap_single(eth->dev, e->dma_addr, e->dma_len, - DMA_TO_DEVICE); - - WRITE_ONCE(desc->msg0, 0); - WRITE_ONCE(desc->msg1, 0); + if (!(desc_ctrl & QDMA_DESC_DONE_MASK) && + !(desc_ctrl & QDMA_DESC_DROP_MASK)) + goto unlock; + + e = &q->entry[index]; + skb = e->skb; + + dma_unmap_single(eth->dev, e->dma_addr, e->dma_len, + DMA_TO_DEVICE); + memset(e, 0, sizeof(*e)); + WRITE_ONCE(desc->msg0, 0); + WRITE_ONCE(desc->msg1, 0); + q->queued--; + + /* completion ring can report out-of-order indexes if hw QoS + * is enabled and packets with different priority are queued + * to same DMA ring. Take into account possible out-of-order + * reports incrementing DMA ring tail pointer + */ + while (q->tail != q->head && !q->entry[q->tail].dma_addr) + q->tail = (q->tail + 1) % q->ndesc; - if (skb) { - u16 queue = skb_get_queue_mapping(skb); - struct netdev_queue *txq; - - txq = netdev_get_tx_queue(skb->dev, queue); - netdev_tx_completed_queue(txq, 1, skb->len); - if (netif_tx_queue_stopped(txq) && - q->ndesc - q->queued >= q->free_thr) - netif_tx_wake_queue(txq); - - dev_kfree_skb_any(skb); - e->skb = NULL; - } + if (skb) { + u16 queue = skb_get_queue_mapping(skb); + struct netdev_queue *txq; + + txq = netdev_get_tx_queue(skb->dev, queue); + netdev_tx_completed_queue(txq, 1, skb->len); + if (netif_tx_queue_stopped(txq) && + q->ndesc - q->queued >= q->free_thr) + netif_tx_wake_queue(txq); - if (index == last) - break; + dev_kfree_skb_any(skb); } - +unlock: spin_unlock_bh(&q->lock); }