realtek: enable CRC offloading

Enables CRC calculation offloading on RTL8380/8390/9300.

Tested on Zyxel XGS1210-10 (RTL9302)/GS1900-48 (RTL8390)/GS1900-10HP (RTL8382)
On the Zyxel GS1900-10HP, an increase of 5% in iperf3 send throughput
and 11% in receive throughput is seen.

Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
This commit is contained in:
Birger Koblitz 2021-04-29 22:53:59 +02:00 committed by Petr Štetiar
parent 3bae740311
commit 9e8d62e421
1 changed files with 96 additions and 27 deletions

View File

@ -234,19 +234,21 @@ struct dsa_tag {
u16 port; u16 port;
u8 l2_offloaded; u8 l2_offloaded;
u8 prio; u8 prio;
bool crc_error;
}; };
bool rtl838x_decode_tag(struct p_hdr *h, struct dsa_tag *t) bool rtl838x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
{ {
t->reason = h->cpu_tag[3] & 0xf; t->reason = h->cpu_tag[3] & 0xf;
if (t->reason != 15)
pr_debug("Reason: %d\n", t->reason);
t->queue = (h->cpu_tag[0] & 0xe0) >> 5; t->queue = (h->cpu_tag[0] & 0xe0) >> 5;
t->port = h->cpu_tag[1] & 0x1f;
t->crc_error = t->reason == 13;
pr_debug("Reason: %d\n", t->reason);
if (t->reason != 4) // NIC_RX_REASON_SPECIAL_TRAP if (t->reason != 4) // NIC_RX_REASON_SPECIAL_TRAP
t->l2_offloaded = 1; t->l2_offloaded = 1;
else else
t->l2_offloaded = 0; t->l2_offloaded = 0;
t->port = h->cpu_tag[1] & 0x1f;
return t->l2_offloaded; return t->l2_offloaded;
} }
@ -254,15 +256,31 @@ bool rtl838x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
bool rtl839x_decode_tag(struct p_hdr *h, struct dsa_tag *t) bool rtl839x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
{ {
t->reason = h->cpu_tag[4] & 0x1f; t->reason = h->cpu_tag[4] & 0x1f;
if (t->reason != 31) t->queue = (h->cpu_tag[3] & 0xe000) >> 13;
t->port = h->cpu_tag[1] & 0x3f;
t->crc_error = h->cpu_tag[3] & BIT(2);
pr_debug("Reason: %d\n", t->reason); pr_debug("Reason: %d\n", t->reason);
t->queue = (h->cpu_tag[3] & 0xe000) >> 13;
if ((t->reason != 7) && (t->reason != 8)) // NIC_RX_REASON_RMA_USR if ((t->reason != 7) && (t->reason != 8)) // NIC_RX_REASON_RMA_USR
t->l2_offloaded = 1; t->l2_offloaded = 1;
else else
t->l2_offloaded = 0; t->l2_offloaded = 0;
t->port = h->cpu_tag[1] & 0x3f; return t->l2_offloaded;
}
bool rtl930x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
{
t->reason = h->cpu_tag[7] & 0x3f;
t->queue = (h->cpu_tag[2] >> 11) & 0x1f;
t->port = (h->cpu_tag[0] >> 8) & 0x1f;
t->crc_error = h->cpu_tag[1] & BIT(6);
pr_debug("Reason %d, port %d, queue %d\n", t->reason, t->port, t->queue);
if (t->reason >= 19 && t->reason <= 27)
t->l2_offloaded = 0;
else
t->l2_offloaded = 1;
return t->l2_offloaded; return t->l2_offloaded;
} }
@ -270,24 +288,19 @@ bool rtl839x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
bool rtl931x_decode_tag(struct p_hdr *h, struct dsa_tag *t) bool rtl931x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
{ {
t->reason = h->cpu_tag[7] & 0x3f; t->reason = h->cpu_tag[7] & 0x3f;
pr_debug("Reason %d\n", t->reason);
t->queue = (h->cpu_tag[2] >> 11) & 0x1f; t->queue = (h->cpu_tag[2] >> 11) & 0x1f;
t->port = (h->cpu_tag[0] >> 8) & 0x3f;
t->crc_error = h->cpu_tag[1] & BIT(6);
pr_debug("Reason %d, port %d, queue %d\n", t->reason, t->port, t->queue);
if (t->reason >= 19 && t->reason <= 27) if (t->reason >= 19 && t->reason <= 27)
t->l2_offloaded = 0; t->l2_offloaded = 0;
else else
t->l2_offloaded = 1; t->l2_offloaded = 1;
t->port = (h->cpu_tag[0] >> 8) & 0x3f;
return t->l2_offloaded; return t->l2_offloaded;
} }
bool rtl930x_decode_tag(struct p_hdr *h, struct dsa_tag *t)
{
rtl931x_decode_tag(h, t);
t->port &= 0x1f;
return t->l2_offloaded;
}
/* /*
* Discard the RX ring-buffers, called as part of the net-ISR * Discard the RX ring-buffers, called as part of the net-ISR
* when the buffer runs over * when the buffer runs over
@ -731,8 +744,9 @@ static void rtl838x_hw_en_rxtx(struct rtl838x_eth_priv *priv)
* | MEDIA_SEL * | MEDIA_SEL
*/ */
sw_w32(0x6192F, priv->r->mac_force_mode_ctrl + priv->cpu_port * 4); sw_w32(0x6192F, priv->r->mac_force_mode_ctrl + priv->cpu_port * 4);
/* allow CRC errors on CPU-port */
sw_w32_mask(0, 0x8, priv->r->mac_port_ctrl(priv->cpu_port)); /* Enable CRC checks on CPU-port */
sw_w32_mask(0, BIT(3), priv->r->mac_port_ctrl(priv->cpu_port));
} }
static void rtl839x_hw_en_rxtx(struct rtl838x_eth_priv *priv) static void rtl839x_hw_en_rxtx(struct rtl838x_eth_priv *priv)
@ -746,8 +760,8 @@ static void rtl839x_hw_en_rxtx(struct rtl838x_eth_priv *priv)
/* Enable DMA */ /* Enable DMA */
sw_w32_mask(0, RX_EN | TX_EN, priv->r->dma_if_ctrl); sw_w32_mask(0, RX_EN | TX_EN, priv->r->dma_if_ctrl);
/* Restart TX/RX to CPU port */ /* Restart TX/RX to CPU port, enable CRC checking */
sw_w32_mask(0x0, 0x3, priv->r->mac_port_ctrl(priv->cpu_port)); sw_w32_mask(0x0, 0x3 | BIT(3), priv->r->mac_port_ctrl(priv->cpu_port));
/* CPU port joins Lookup Miss Flooding Portmask */ /* CPU port joins Lookup Miss Flooding Portmask */
// TODO: The code below should also work for the RTL838x // TODO: The code below should also work for the RTL838x
@ -784,8 +798,8 @@ static void rtl93xx_hw_en_rxtx(struct rtl838x_eth_priv *priv)
/* Enable DMA */ /* Enable DMA */
sw_w32_mask(0, RX_EN_93XX | TX_EN_93XX, priv->r->dma_if_ctrl); sw_w32_mask(0, RX_EN_93XX | TX_EN_93XX, priv->r->dma_if_ctrl);
/* Restart TX/RX to CPU port */ /* Restart TX/RX to CPU port, enable CRC checking */
sw_w32_mask(0x0, 0x3, priv->r->mac_port_ctrl(priv->cpu_port)); sw_w32_mask(0x0, 0x3 | BIT(4), priv->r->mac_port_ctrl(priv->cpu_port));
sw_w32_mask(0, BIT(priv->cpu_port), RTL930X_L2_UNKN_UC_FLD_PMSK); sw_w32_mask(0, BIT(priv->cpu_port), RTL930X_L2_UNKN_UC_FLD_PMSK);
sw_w32(0x217, priv->r->mac_force_mode_ctrl + priv->cpu_port * 4); sw_w32(0x217, priv->r->mac_force_mode_ctrl + priv->cpu_port * 4);
@ -1115,15 +1129,20 @@ static int rtl838x_eth_tx(struct sk_buff *skb, struct net_device *dev)
if (netdev_uses_dsa(dev) && skb->data[len-4] == 0x80 && skb->data[len-3] > 0 if (netdev_uses_dsa(dev) && skb->data[len-4] == 0x80 && skb->data[len-3] > 0
&& skb->data[len-3] < 28 && skb->data[len-2] == 0x10 && skb->data[len-3] < 28 && skb->data[len-2] == 0x10
&& skb->data[len-1] == 0x00) { && skb->data[len-1] == 0x00) {
/* Reuse tag space for CRC */ /* Reuse tag space for CRC if possible */
dest_port = skb->data[len-3]; dest_port = skb->data[len-3];
len -= 4; len -= 4;
} }
if (len < ETH_ZLEN)
len = ETH_ZLEN;
/* ASIC expects that packet includes CRC, so we extend by 4 bytes */ len += 4; // Add space for CRC
len += 4;
// On RTL8380 SoCs, the packet needs extra padding
if (priv->family_id == RTL8380_FAMILY_ID) {
if (len < ETH_ZLEN)
len = ETH_ZLEN; // SoC not automatically padding to ETH_ZLEN
else
len += 4;
}
if (skb_padto(skb, len)) { if (skb_padto(skb, len)) {
ret = NETDEV_TX_OK; ret = NETDEV_TX_OK;
@ -1280,6 +1299,12 @@ static int rtl838x_hw_receive(struct net_device *dev, int r, int budget)
tag.queue, len, tag.reason, tag.port); tag.queue, len, tag.reason, tag.port);
skb->protocol = eth_type_trans(skb, dev); skb->protocol = eth_type_trans(skb, dev);
if (dev->features & NETIF_F_RXCSUM) {
if (tag.crc_error)
skb_checksum_none_assert(skb);
else
skb->ip_summed = CHECKSUM_UNNECESSARY;
}
dev->stats.rx_packets++; dev->stats.rx_packets++;
dev->stats.rx_bytes += len; dev->stats.rx_bytes += len;
@ -1837,6 +1862,40 @@ static int rtl838x_mdio_remove(struct rtl838x_eth_priv *priv)
return 0; return 0;
} }
static netdev_features_t rtl838x_fix_features(struct net_device *dev,
netdev_features_t features)
{
return features;
}
static int rtl83xx_set_features(struct net_device *dev, netdev_features_t features)
{
struct rtl838x_eth_priv *priv = netdev_priv(dev);
if ((features ^ dev->features) & NETIF_F_RXCSUM) {
if (!(features & NETIF_F_RXCSUM))
sw_w32_mask(BIT(3), 0, priv->r->mac_port_ctrl(priv->cpu_port));
else
sw_w32_mask(0, BIT(4), priv->r->mac_port_ctrl(priv->cpu_port));
}
return 0;
}
static int rtl93xx_set_features(struct net_device *dev, netdev_features_t features)
{
struct rtl838x_eth_priv *priv = netdev_priv(dev);
if ((features ^ dev->features) & NETIF_F_RXCSUM) {
if (!(features & NETIF_F_RXCSUM))
sw_w32_mask(BIT(4), 0, priv->r->mac_port_ctrl(priv->cpu_port));
else
sw_w32_mask(0, BIT(4), priv->r->mac_port_ctrl(priv->cpu_port));
}
return 0;
}
static const struct net_device_ops rtl838x_eth_netdev_ops = { static const struct net_device_ops rtl838x_eth_netdev_ops = {
.ndo_open = rtl838x_eth_open, .ndo_open = rtl838x_eth_open,
.ndo_stop = rtl838x_eth_stop, .ndo_stop = rtl838x_eth_stop,
@ -1846,6 +1905,8 @@ static const struct net_device_ops rtl838x_eth_netdev_ops = {
.ndo_validate_addr = eth_validate_addr, .ndo_validate_addr = eth_validate_addr,
.ndo_set_rx_mode = rtl838x_eth_set_multicast_list, .ndo_set_rx_mode = rtl838x_eth_set_multicast_list,
.ndo_tx_timeout = rtl838x_eth_tx_timeout, .ndo_tx_timeout = rtl838x_eth_tx_timeout,
.ndo_set_features = rtl83xx_set_features,
.ndo_fix_features = rtl838x_fix_features,
}; };
static const struct net_device_ops rtl839x_eth_netdev_ops = { static const struct net_device_ops rtl839x_eth_netdev_ops = {
@ -1857,6 +1918,8 @@ static const struct net_device_ops rtl839x_eth_netdev_ops = {
.ndo_validate_addr = eth_validate_addr, .ndo_validate_addr = eth_validate_addr,
.ndo_set_rx_mode = rtl839x_eth_set_multicast_list, .ndo_set_rx_mode = rtl839x_eth_set_multicast_list,
.ndo_tx_timeout = rtl838x_eth_tx_timeout, .ndo_tx_timeout = rtl838x_eth_tx_timeout,
.ndo_set_features = rtl83xx_set_features,
.ndo_fix_features = rtl838x_fix_features,
}; };
static const struct net_device_ops rtl930x_eth_netdev_ops = { static const struct net_device_ops rtl930x_eth_netdev_ops = {
@ -1868,6 +1931,8 @@ static const struct net_device_ops rtl930x_eth_netdev_ops = {
.ndo_validate_addr = eth_validate_addr, .ndo_validate_addr = eth_validate_addr,
.ndo_set_rx_mode = rtl930x_eth_set_multicast_list, .ndo_set_rx_mode = rtl930x_eth_set_multicast_list,
.ndo_tx_timeout = rtl838x_eth_tx_timeout, .ndo_tx_timeout = rtl838x_eth_tx_timeout,
.ndo_set_features = rtl93xx_set_features,
.ndo_fix_features = rtl838x_fix_features,
}; };
static const struct net_device_ops rtl931x_eth_netdev_ops = { static const struct net_device_ops rtl931x_eth_netdev_ops = {
@ -1879,6 +1944,8 @@ static const struct net_device_ops rtl931x_eth_netdev_ops = {
.ndo_validate_addr = eth_validate_addr, .ndo_validate_addr = eth_validate_addr,
.ndo_set_rx_mode = rtl931x_eth_set_multicast_list, .ndo_set_rx_mode = rtl931x_eth_set_multicast_list,
.ndo_tx_timeout = rtl838x_eth_tx_timeout, .ndo_tx_timeout = rtl838x_eth_tx_timeout,
.ndo_set_features = rtl93xx_set_features,
.ndo_fix_features = rtl838x_fix_features,
}; };
static const struct phylink_mac_ops rtl838x_phylink_ops = { static const struct phylink_mac_ops rtl838x_phylink_ops = {
@ -1975,6 +2042,8 @@ static int __init rtl838x_eth_probe(struct platform_device *pdev)
dev->ethtool_ops = &rtl838x_ethtool_ops; dev->ethtool_ops = &rtl838x_ethtool_ops;
dev->min_mtu = ETH_ZLEN; dev->min_mtu = ETH_ZLEN;
dev->max_mtu = 1536; dev->max_mtu = 1536;
dev->features = NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
dev->hw_features = NETIF_F_RXCSUM;
priv->id = soc_info.id; priv->id = soc_info.id;
priv->family_id = soc_info.family; priv->family_id = soc_info.family;