Implement .ndo_start_xmit function to fill the information of the packet
to be transmitted into the tx descriptor, and then the hardware will
transmit the packet using the information in the tx descriptor.
In addition, we also implemented the tx_handler function to enable the
tx descriptor to be reused.
Signed-off-by: Justin Lai <justinlai0215@realtek.com>
---
.../net/ethernet/realtek/rtase/rtase_main.c | 285 ++++++++++++++++++
1 file changed, 285 insertions(+)
diff --git a/drivers/net/ethernet/realtek/rtase/rtase_main.c b/drivers/net/ethernet/realtek/rtase/rtase_main.c
index b6304fd3f907..54bf683ae50e 100644
--- a/drivers/net/ethernet/realtek/rtase/rtase_main.c
+++ b/drivers/net/ethernet/realtek/rtase/rtase_main.c
@@ -253,6 +253,68 @@ static void rtase_mark_to_asic(union rtase_rx_desc *desc, u32 rx_buf_sz)
cpu_to_le32(RTASE_DESC_OWN | eor | rx_buf_sz));
}
+static u32 rtase_tx_avail(struct rtase_ring *ring)
+{
+ return READ_ONCE(ring->dirty_idx) + RTASE_NUM_DESC -
+ READ_ONCE(ring->cur_idx);
+}
+
+static int tx_handler(struct rtase_ring *ring, int budget)
+{
+ const struct rtase_private *tp = ring->ivec->tp;
+ struct net_device *dev = tp->dev;
+ u32 dirty_tx, tx_left;
+ u32 bytes_compl = 0;
+ u32 pkts_compl = 0;
+ int workdone = 0;
+
+ dirty_tx = ring->dirty_idx;
+ tx_left = READ_ONCE(ring->cur_idx) - dirty_tx;
+
+ while (tx_left > 0) {
+ u32 entry = dirty_tx % RTASE_NUM_DESC;
+ struct rtase_tx_desc *desc = ring->desc +
+ sizeof(struct rtase_tx_desc) * entry;
+ u32 status;
+
+ status = le32_to_cpu(desc->opts1);
+
+ if (status & RTASE_DESC_OWN)
+ break;
+
+ rtase_unmap_tx_skb(tp->pdev, ring->mis.len[entry], desc);
+ ring->mis.len[entry] = 0;
+ if (ring->skbuff[entry]) {
+ pkts_compl++;
+ bytes_compl += ring->skbuff[entry]->len;
+ napi_consume_skb(ring->skbuff[entry], budget);
+ ring->skbuff[entry] = NULL;
+ }
+
+ dirty_tx++;
+ tx_left--;
+ workdone++;
+
+ if (workdone == RTASE_TX_BUDGET_DEFAULT)
+ break;
+ }
+
+ if (ring->dirty_idx != dirty_tx) {
+ dev_sw_netstats_tx_add(dev, pkts_compl, bytes_compl);
+ WRITE_ONCE(ring->dirty_idx, dirty_tx);
+
+ netif_subqueue_completed_wake(dev, ring->index, pkts_compl,
+ bytes_compl,
+ rtase_tx_avail(ring),
+ RTASE_TX_START_THRS);
+
+ if (ring->cur_idx != dirty_tx)
+ rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index));
+ }
+
+ return 0;
+}
+
static void rtase_tx_desc_init(struct rtase_private *tp, u16 idx)
{
struct rtase_ring *ring = &tp->tx_ring[idx];
@@ -1004,6 +1066,228 @@ static int rtase_close(struct net_device *dev)
return 0;
}
+static u32 rtase_tx_vlan_tag(const struct rtase_private *tp,
+ const struct sk_buff *skb)
+{
+ return (skb_vlan_tag_present(skb)) ?
+ (RTASE_TX_VLAN_TAG | swab16(skb_vlan_tag_get(skb))) : 0x00;
+}
+
+static u32 rtase_tx_csum(struct sk_buff *skb, const struct net_device *dev)
+{
+ u32 csum_cmd = 0;
+ u8 ip_protocol;
+
+ switch (vlan_get_protocol(skb)) {
+ case htons(ETH_P_IP):
+ csum_cmd = RTASE_TX_IPCS_C;
+ ip_protocol = ip_hdr(skb)->protocol;
+ break;
+
+ case htons(ETH_P_IPV6):
+ csum_cmd = RTASE_TX_IPV6F_C;
+ ip_protocol = ipv6_hdr(skb)->nexthdr;
+ break;
+
+ default:
+ ip_protocol = IPPROTO_RAW;
+ break;
+ }
+
+ if (ip_protocol == IPPROTO_TCP)
+ csum_cmd |= RTASE_TX_TCPCS_C;
+ else if (ip_protocol == IPPROTO_UDP)
+ csum_cmd |= RTASE_TX_UDPCS_C;
+
+ csum_cmd |= u32_encode_bits(skb_transport_offset(skb),
+ RTASE_TCPHO_MASK);
+
+ return csum_cmd;
+}
+
+static int rtase_xmit_frags(struct rtase_ring *ring, struct sk_buff *skb,
+ u32 opts1, u32 opts2)
+{
+ const struct skb_shared_info *info = skb_shinfo(skb);
+ const struct rtase_private *tp = ring->ivec->tp;
+ const u8 nr_frags = info->nr_frags;
+ struct rtase_tx_desc *txd = NULL;
+ u32 cur_frag, entry;
+
+ entry = ring->cur_idx;
+ for (cur_frag = 0; cur_frag < nr_frags; cur_frag++) {
+ const skb_frag_t *frag = &info->frags[cur_frag];
+ dma_addr_t mapping;
+ u32 status, len;
+ void *addr;
+
+ entry = (entry + 1) % RTASE_NUM_DESC;
+
+ txd = ring->desc + sizeof(struct rtase_tx_desc) * entry;
+ len = skb_frag_size(frag);
+ addr = skb_frag_address(frag);
+ mapping = dma_map_single(&tp->pdev->dev, addr, len,
+ DMA_TO_DEVICE);
+
+ if (unlikely(dma_mapping_error(&tp->pdev->dev, mapping))) {
+ if (unlikely(net_ratelimit()))
+ netdev_err(tp->dev,
+ "Failed to map TX fragments DMA!\n");
+
+ goto err_out;
+ }
+
+ if (((entry + 1) % RTASE_NUM_DESC) == 0)
+ status = (opts1 | len | RTASE_RING_END);
+ else
+ status = opts1 | len;
+
+ if (cur_frag == (nr_frags - 1)) {
+ ring->skbuff[entry] = skb;
+ status |= RTASE_TX_LAST_FRAG;
+ }
+
+ ring->mis.len[entry] = len;
+ txd->addr = cpu_to_le64(mapping);
+ txd->opts2 = cpu_to_le32(opts2);
+
+ /* make sure the operating fields have been updated */
+ dma_wmb();
+ txd->opts1 = cpu_to_le32(status);
+ }
+
+ return cur_frag;
+
+err_out:
+ rtase_tx_clear_range(ring, ring->cur_idx + 1, cur_frag);
+ return -EIO;
+}
+
+static netdev_tx_t rtase_start_xmit(struct sk_buff *skb,
+ struct net_device *dev)
+{
+ struct skb_shared_info *shinfo = skb_shinfo(skb);
+ struct rtase_private *tp = netdev_priv(dev);
+ u32 q_idx, entry, len, opts1, opts2;
+ struct netdev_queue *tx_queue;
+ bool stop_queue, door_bell;
+ u32 mss = shinfo->gso_size;
+ struct rtase_tx_desc *txd;
+ struct rtase_ring *ring;
+ dma_addr_t mapping;
+ int frags;
+
+ /* multiqueues */
+ q_idx = skb_get_queue_mapping(skb);
+ ring = &tp->tx_ring[q_idx];
+ tx_queue = netdev_get_tx_queue(dev, q_idx);
+
+ if (unlikely(!rtase_tx_avail(ring))) {
+ if (net_ratelimit())
+ netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
+ goto err_stop;
+ }
+
+ entry = ring->cur_idx % RTASE_NUM_DESC;
+ txd = ring->desc + sizeof(struct rtase_tx_desc) * entry;
+
+ opts1 = RTASE_DESC_OWN;
+ opts2 = rtase_tx_vlan_tag(tp, skb);
+
+ /* tcp segmentation offload (or tcp large send) */
+ if (mss) {
+ if (shinfo->gso_type & SKB_GSO_TCPV4) {
+ opts1 |= RTASE_GIANT_SEND_V4;
+ } else if (shinfo->gso_type & SKB_GSO_TCPV6) {
+ if (skb_cow_head(skb, 0))
+ goto err_dma_0;
+
+ tcp_v6_gso_csum_prep(skb);
+ opts1 |= RTASE_GIANT_SEND_V6;
+ } else {
+ WARN_ON_ONCE(1);
+ }
+
+ opts1 |= u32_encode_bits(skb_transport_offset(skb),
+ RTASE_TCPHO_MASK);
+ opts2 |= u32_encode_bits(mss, RTASE_MSS_MASK);
+ } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
+ opts2 |= rtase_tx_csum(skb, dev);
+ }
+
+ frags = rtase_xmit_frags(ring, skb, opts1, opts2);
+ if (unlikely(frags < 0))
+ goto err_dma_0;
+
+ if (frags) {
+ len = skb_headlen(skb);
+ opts1 |= RTASE_TX_FIRST_FRAG;
+ } else {
+ len = skb->len;
+ ring->skbuff[entry] = skb;
+ opts1 |= RTASE_TX_FIRST_FRAG | RTASE_TX_LAST_FRAG;
+ }
+
+ if (((entry + 1) % RTASE_NUM_DESC) == 0)
+ opts1 |= (len | RTASE_RING_END);
+ else
+ opts1 |= len;
+
+ mapping = dma_map_single(&tp->pdev->dev, skb->data, len,
+ DMA_TO_DEVICE);
+
+ if (unlikely(dma_mapping_error(&tp->pdev->dev, mapping))) {
+ if (unlikely(net_ratelimit()))
+ netdev_err(dev, "Failed to map TX DMA!\n");
+
+ goto err_dma_1;
+ }
+
+ ring->mis.len[entry] = len;
+ txd->addr = cpu_to_le64(mapping);
+ txd->opts2 = cpu_to_le32(opts2);
+ txd->opts1 = cpu_to_le32(opts1 & ~RTASE_DESC_OWN);
+
+ /* make sure the operating fields have been updated */
+ dma_wmb();
+
+ door_bell = __netdev_tx_sent_queue(tx_queue, skb->len,
+ netdev_xmit_more());
+
+ txd->opts1 = cpu_to_le32(opts1);
+
+ skb_tx_timestamp(skb);
+
+ /* tx needs to see descriptor changes before updated cur_idx */
+ smp_wmb();
+
+ WRITE_ONCE(ring->cur_idx, ring->cur_idx + frags + 1);
+
+ stop_queue = !netif_subqueue_maybe_stop(dev, ring->index,
+ rtase_tx_avail(ring),
+ RTASE_TX_STOP_THRS,
+ RTASE_TX_START_THRS);
+
+ if (door_bell || stop_queue)
+ rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index));
+
+ return NETDEV_TX_OK;
+
+err_dma_1:
+ ring->skbuff[entry] = NULL;
+ rtase_tx_clear_range(ring, ring->cur_idx + 1, frags);
+
+err_dma_0:
+ tp->stats.tx_dropped++;
+ dev_kfree_skb_any(skb);
+ return NETDEV_TX_OK;
+
+err_stop:
+ netif_stop_queue(dev);
+ tp->stats.tx_dropped++;
+ return NETDEV_TX_BUSY;
+}
+
static void rtase_enable_eem_write(const struct rtase_private *tp)
{
u8 val;
@@ -1039,6 +1323,7 @@ static void rtase_rar_set(const struct rtase_private *tp, const u8 *addr)
static const struct net_device_ops rtase_netdev_ops = {
.ndo_open = rtase_open,
.ndo_stop = rtase_close,
+ .ndo_start_xmit = rtase_start_xmit,
};
static void rtase_get_mac_address(struct net_device *dev)
--
2.34.1On Mon, 29 Jul 2024 14:21:14 +0800 Justin Lai wrote: > + stop_queue = !netif_subqueue_maybe_stop(dev, ring->index, > + rtase_tx_avail(ring), > + RTASE_TX_STOP_THRS, > + RTASE_TX_START_THRS); > + > + if (door_bell || stop_queue) > + rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index)); > + > + return NETDEV_TX_OK; > + > +err_dma_1: > + ring->skbuff[entry] = NULL; > + rtase_tx_clear_range(ring, ring->cur_idx + 1, frags); > + > +err_dma_0: > + tp->stats.tx_dropped++; > + dev_kfree_skb_any(skb); > + return NETDEV_TX_OK; > + > +err_stop: > + netif_stop_queue(dev); > + tp->stats.tx_dropped++; > + return NETDEV_TX_BUSY; If you're dropping a packet you should somehow check that the previous xmit didn't enqueue a packet to the ring and skip the doorbell because door_bell was false. If that's the case you have to ring the doorbell now. Also you shouldn't increment dropped if you return TX_BUSY. Please fix these issues in the driver you're copying from, too.
> On Mon, 29 Jul 2024 14:21:14 +0800 Justin Lai wrote: > > + stop_queue = !netif_subqueue_maybe_stop(dev, ring->index, > > + rtase_tx_avail(ring), > > + > RTASE_TX_STOP_THRS, > > + > RTASE_TX_START_THRS); > > + > > + if (door_bell || stop_queue) > > + rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index)); > > + > > + return NETDEV_TX_OK; > > + > > +err_dma_1: > > + ring->skbuff[entry] = NULL; > > + rtase_tx_clear_range(ring, ring->cur_idx + 1, frags); > > + > > +err_dma_0: > > + tp->stats.tx_dropped++; > > + dev_kfree_skb_any(skb); > > + return NETDEV_TX_OK; > > + > > +err_stop: > > + netif_stop_queue(dev); > > + tp->stats.tx_dropped++; > > + return NETDEV_TX_BUSY; > > If you're dropping a packet you should somehow check that the previous xmit > didn't enqueue a packet to the ring and skip the doorbell because door_bell > was false. If that's the case you have to ring the doorbell now. > > Also you shouldn't increment dropped if you return TX_BUSY. Thank you for your response. I will modify it. > > Please fix these issues in the driver you're copying from, too. I will inform the person responsible for the R8169 of the issues we've identified and discuss what modifications need to be made. Thanks Justin
On 30.07.2024 11:27, Justin Lai wrote: >> On Mon, 29 Jul 2024 14:21:14 +0800 Justin Lai wrote: >>> + stop_queue = !netif_subqueue_maybe_stop(dev, ring->index, >>> + rtase_tx_avail(ring), >>> + >> RTASE_TX_STOP_THRS, >>> + >> RTASE_TX_START_THRS); >>> + >>> + if (door_bell || stop_queue) >>> + rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index)); >>> + >>> + return NETDEV_TX_OK; >>> + >>> +err_dma_1: >>> + ring->skbuff[entry] = NULL; >>> + rtase_tx_clear_range(ring, ring->cur_idx + 1, frags); >>> + >>> +err_dma_0: >>> + tp->stats.tx_dropped++; >>> + dev_kfree_skb_any(skb); >>> + return NETDEV_TX_OK; >>> + >>> +err_stop: >>> + netif_stop_queue(dev); >>> + tp->stats.tx_dropped++; >>> + return NETDEV_TX_BUSY; >> >> If you're dropping a packet you should somehow check that the previous xmit >> didn't enqueue a packet to the ring and skip the doorbell because door_bell >> was false. If that's the case you have to ring the doorbell now. >> I briefly checked and this may be a common issue, e.g. stmmac_xmit() behaves the same as r8169 here. Hitting this error path should be a rare event, so we could ring the doorbell always, w/o any further checks? >> Also you shouldn't increment dropped if you return TX_BUSY. > I'll take care of this change for r8169. > Thank you for your response. I will modify it. >> >> Please fix these issues in the driver you're copying from, too. > > I will inform the person responsible for the R8169 of the issues we've > identified and discuss what modifications need to be made. > > Thanks > Justin >
On Tue, 30 Jul 2024 09:27:27 +0000 Justin Lai wrote: > > Please fix these issues in the driver you're copying from, too. > > I will inform the person responsible for the R8169 of the issues we've > identified and discuss what modifications need to be made. That sounds quite noncommittal. Please fix r8169 _before_ you post the next version of this cop^W driver.
© 2016 - 2026 Red Hat, Inc.