[PATCH net-next v2 0/5] atl1c: support for Mikrotik 10/25G NIC features

[PATCH net-next v2 0/5] atl1c: support for Mikrotik 10/25G NIC features

[Gatis Peisenieks]

The new Mikrotik 10/25G NIC maintains compatibility with existing atl1c
driver. However it does have new features.

This patch set adds support for reporting cards higher link speed, max-mtu,
enables rx csum offload and improves tx performance.

v2:
    - fixed xmit_more handling as pointed out by Eric Dumazet
    - added a more reliable link detection on Mikrotik 10/25G NIC
      since MDIO op emulation can occasionally fail

Gatis Peisenieks (5):
  atl1c: show correct link speed on Mikrotik 10/25G NIC
  atl1c: improve performance by avoiding unnecessary pcie writes on xmit
  atl1c: adjust max mtu according to Mikrotik 10/25G NIC ability
  atl1c: enable rx csum offload on Mikrotik 10/25G NIC
  atl1c: improve link detection reliability on Mikrotik 10/25G NIC

 drivers/net/ethernet/atheros/atl1c/atl1c.h    |  3 +
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.c | 35 +++++++++---
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.h |  8 +++
 .../net/ethernet/atheros/atl1c/atl1c_main.c   | 55 ++++++++++++-------
 4 files changed, 75 insertions(+), 26 deletions(-)


base-commit: d8654f4f9300e5e7cf8d5e7885978541cf61326b
-- 
2.31.1

[PATCH net-next v2 1/5] atl1c: show correct link speed on Mikrotik 10/25G NIC

[Gatis Peisenieks]

The new Mikrotik 10/25G NIC maintains compatibility with existing atl1c
driver. However it does have new features.

This defines some new register offsets, code for identifying the new type
of NIC and correct speed detection for the NIC.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
 drivers/net/ethernet/atheros/atl1c/atl1c.h      | 1 +
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.c   | 9 +++++++++
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.h   | 7 +++++++
 drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 4 ++++
 4 files changed, 21 insertions(+)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c.h b/drivers/net/ethernet/atheros/atl1c/atl1c.h
index 28ae5c16831e..3fda7eb3bd69 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c.h
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c.h
@@ -289,6 +289,7 @@ enum atl1c_nic_type {
 	athr_l2c_b2,
 	athr_l1d,
 	athr_l1d_2,
+	athr_mt,
 };
 
 enum atl1c_trans_queue {
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
index 140358dcf61e..ddb9442416cd 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
@@ -648,6 +648,15 @@ int atl1c_get_speed_and_duplex(struct atl1c_hw *hw, u16 *speed, u16 *duplex)
 	int err;
 	u16 phy_data;
 
+	if (hw->nic_type == athr_mt) {
+		u32 spd;
+
+		AT_READ_REG(hw, REG_MT_SPEED, &spd);
+		*speed = spd;
+		*duplex = FULL_DUPLEX;
+		return 0;
+	}
+
 	/* Read   PHY Specific Status Register (17) */
 	err = atl1c_read_phy_reg(hw, MII_GIGA_PSSR, &phy_data);
 	if (err)
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
index ce1a123dce2c..73cbc049a63e 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
@@ -764,6 +764,13 @@ void atl1c_post_phy_linkchg(struct atl1c_hw *hw, u16 link_speed);
 #define REG_DEBUG_DATA0 		0x1900
 #define REG_DEBUG_DATA1 		0x1904
 
+#define REG_MT_MAGIC			0x1F00
+#define REG_MT_MODE			0x1F04
+#define REG_MT_SPEED			0x1F08
+#define REG_MT_VERSION			0x1F0C
+
+#define MT_MAGIC			0xaabb1234
+
 #define L1D_MPW_PHYID1			0xD01C  /* V7 */
 #define L1D_MPW_PHYID2			0xD01D  /* V1-V6 */
 #define L1D_MPW_PHYID3			0xD01E  /* V8 */
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index c6263cf8d3c0..28c30d5288e4 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -644,6 +644,7 @@ static int atl1c_alloc_queues(struct atl1c_adapter *adapter)
 
 static void atl1c_set_mac_type(struct atl1c_hw *hw)
 {
+	u32 magic;
 	switch (hw->device_id) {
 	case PCI_DEVICE_ID_ATTANSIC_L2C:
 		hw->nic_type = athr_l2c;
@@ -662,6 +663,9 @@ static void atl1c_set_mac_type(struct atl1c_hw *hw)
 		break;
 	case PCI_DEVICE_ID_ATHEROS_L1D_2_0:
 		hw->nic_type = athr_l1d_2;
+		AT_READ_REG(hw, REG_MT_MAGIC, &magic);
+		if (magic == MT_MAGIC)
+			hw->nic_type = athr_mt;
 		break;
 	default:
 		break;
-- 
2.31.1

[PATCH net-next v2 2/5] atl1c: improve performance by avoiding unnecessary pcie writes on xmit

[Gatis Peisenieks]

The kernel has xmit_more facility that hints the networking driver xmit
path about whether more packets are coming soon. This information can be
used to avoid unnecessary expensive PCIe transaction per tx packet.

Max TX pps on Mikrotik 10/25G NIC in a Threadripper 3960X system
improved from 1150Kpps to 1700Kpps.

Testing L2 forwarding on AR8151 hardware did not reveal a measurable
increase in latency.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
 drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index 28c30d5288e4..08a0f49e03ce 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -2211,8 +2211,8 @@ static int atl1c_tx_map(struct atl1c_adapter *adapter,
 	return -1;
 }
 
-static void atl1c_tx_queue(struct atl1c_adapter *adapter, struct sk_buff *skb,
-			   struct atl1c_tpd_desc *tpd, enum atl1c_trans_queue type)
+static void atl1c_tx_queue(struct atl1c_adapter *adapter,
+			   enum atl1c_trans_queue type)
 {
 	struct atl1c_tpd_ring *tpd_ring = &adapter->tpd_ring[type];
 	u16 reg;
@@ -2238,6 +2238,7 @@ static netdev_tx_t atl1c_xmit_frame(struct sk_buff *skb,
 
 	if (atl1c_tpd_avail(adapter, type) < tpd_req) {
 		/* no enough descriptor, just stop queue */
+		atl1c_tx_queue(adapter, type);
 		netif_stop_queue(netdev);
 		return NETDEV_TX_BUSY;
 	}
@@ -2246,6 +2247,7 @@ static netdev_tx_t atl1c_xmit_frame(struct sk_buff *skb,
 
 	/* do TSO and check sum */
 	if (atl1c_tso_csum(adapter, skb, &tpd, type) != 0) {
+		atl1c_tx_queue(adapter, type);
 		dev_kfree_skb_any(skb);
 		return NETDEV_TX_OK;
 	}
@@ -2270,8 +2272,10 @@ static netdev_tx_t atl1c_xmit_frame(struct sk_buff *skb,
 		atl1c_tx_rollback(adapter, tpd, type);
 		dev_kfree_skb_any(skb);
 	} else {
-		netdev_sent_queue(adapter->netdev, skb->len);
-		atl1c_tx_queue(adapter, skb, tpd, type);
+		bool more = netdev_xmit_more();
+
+		if (__netdev_sent_queue(adapter->netdev, skb->len, more))
+			atl1c_tx_queue(adapter, type);
 	}
 
 	return NETDEV_TX_OK;
-- 
2.31.1

[PATCH net-next v2 3/5] atl1c: adjust max mtu according to Mikrotik 10/25G NIC ability

[Gatis Peisenieks]

The new Mikrotik 10/25G NIC supports jumbo frames. Jumbo frames are
supported for TSO as well.

This enables the support for mtu up to 9500 bytes.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
 drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index 08a0f49e03ce..dbafd8118a86 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -478,6 +478,9 @@ static void atl1c_set_rxbufsize(struct atl1c_adapter *adapter,
 static netdev_features_t atl1c_fix_features(struct net_device *netdev,
 	netdev_features_t features)
 {
+	struct atl1c_adapter *adapter = netdev_priv(netdev);
+	struct atl1c_hw *hw = &adapter->hw;
+
 	/*
 	 * Since there is no support for separate rx/tx vlan accel
 	 * enable/disable make sure tx flag is always in same state as rx.
@@ -487,8 +490,10 @@ static netdev_features_t atl1c_fix_features(struct net_device *netdev,
 	else
 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
 
-	if (netdev->mtu > MAX_TSO_FRAME_SIZE)
-		features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
+	if (hw->nic_type != athr_mt) {
+		if (netdev->mtu > MAX_TSO_FRAME_SIZE)
+			features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
+	}
 
 	return features;
 }
@@ -515,9 +520,12 @@ static void atl1c_set_max_mtu(struct net_device *netdev)
 	case athr_l1d:
 	case athr_l1d_2:
 		netdev->max_mtu = MAX_JUMBO_FRAME_SIZE -
-				  (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
+			(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
+		break;
+	case athr_mt:
+		netdev->max_mtu = 9500;
 		break;
-	/* The 10/100 devices don't support jumbo packets, max_mtu 1500 */
+		/* The 10/100 devices don't support jumbo packets, max_mtu 1500 */
 	default:
 		netdev->max_mtu = ETH_DATA_LEN;
 		break;
-- 
2.31.1

[PATCH net-next v2 4/5] atl1c: enable rx csum offload on Mikrotik 10/25G NIC

[Gatis Peisenieks]

Mikrotik 10/25G NIC supports hw checksum verification on rx for
IP/IPv6 + TCP/UDP packets. HW checksum offload helps reduce host
cpu load.

This enables the csum offload specifically for Mikrotik 10/25G NIC
as other HW supported by the driver is known to have problems with it.

TCP iperf3 to Threadripper 3960X with NIC improved 16.5 -> 20.0 Gbps
with mtu=1500.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
 drivers/net/ethernet/atheros/atl1c/atl1c.h      | 2 ++
 drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 5 +++++
 2 files changed, 7 insertions(+)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c.h b/drivers/net/ethernet/atheros/atl1c/atl1c.h
index 3fda7eb3bd69..9d70cb7544f1 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c.h
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c.h
@@ -241,6 +241,8 @@ struct atl1c_tpd_ext_desc {
 #define RRS_PACKET_PROT_IS_IPV6_ONLY(word) \
 	((((word) >> RRS_PROT_ID_SHIFT) & RRS_PROT_ID_MASK) == 6)
 
+#define RRS_MT_PROT_ID_TCPUDP	BIT(19)
+
 struct atl1c_recv_ret_status {
 	__le32  word0;
 	__le32	rss_hash;
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index dbafd8118a86..9693da5028cf 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -1671,6 +1671,11 @@ static irqreturn_t atl1c_intr(int irq, void *data)
 static inline void atl1c_rx_checksum(struct atl1c_adapter *adapter,
 		  struct sk_buff *skb, struct atl1c_recv_ret_status *prrs)
 {
+	if (adapter->hw.nic_type == athr_mt) {
+		if (prrs->word3 & RRS_MT_PROT_ID_TCPUDP)
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+		return;
+	}
 	/*
 	 * The pid field in RRS in not correct sometimes, so we
 	 * cannot figure out if the packet is fragmented or not,
-- 
2.31.1

[PATCH net-next v2 5/5] atl1c: improve link detection reliability on Mikrotik 10/25G NIC

[Gatis Peisenieks]

Mikrotik 10/25G NIC emulates the MDIO accesses, but the emulation is
not 100% reliable - the MDIO ops occasionally can timeout.

This adds a reliable way of detecting link on Mikrotik 10/25G NIC.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.c | 26 ++++++++++++++-----
 drivers/net/ethernet/atheros/atl1c/atl1c_hw.h |  1 +
 .../net/ethernet/atheros/atl1c/atl1c_main.c   | 18 +++++--------
 3 files changed, 27 insertions(+), 18 deletions(-)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
index ddb9442416cd..7dff20350865 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.c
@@ -636,6 +636,23 @@ int atl1c_phy_init(struct atl1c_hw *hw)
 	return 0;
 }
 
+bool atl1c_get_link_status(struct atl1c_hw *hw)
+{
+	u16 phy_data;
+
+	if (hw->nic_type == athr_mt) {
+		u32 spd;
+
+		AT_READ_REG(hw, REG_MT_SPEED, &spd);
+		return !!spd;
+	}
+
+	/* MII_BMSR must be read twice */
+	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
+	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
+	return !!(phy_data & BMSR_LSTATUS);
+}
+
 /*
  * Detects the current speed and duplex settings of the hardware.
  *
@@ -695,15 +712,12 @@ int atl1c_phy_to_ps_link(struct atl1c_hw *hw)
 	int ret = 0;
 	u16 autoneg_advertised = ADVERTISED_10baseT_Half;
 	u16 save_autoneg_advertised;
-	u16 phy_data;
 	u16 mii_lpa_data;
 	u16 speed = SPEED_0;
 	u16 duplex = FULL_DUPLEX;
 	int i;
 
-	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
-	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
-	if (phy_data & BMSR_LSTATUS) {
+	if (atl1c_get_link_status(hw)) {
 		atl1c_read_phy_reg(hw, MII_LPA, &mii_lpa_data);
 		if (mii_lpa_data & LPA_10FULL)
 			autoneg_advertised = ADVERTISED_10baseT_Full;
@@ -726,9 +740,7 @@ int atl1c_phy_to_ps_link(struct atl1c_hw *hw)
 		if (mii_lpa_data) {
 			for (i = 0; i < AT_SUSPEND_LINK_TIMEOUT; i++) {
 				mdelay(100);
-				atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
-				atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
-				if (phy_data & BMSR_LSTATUS) {
+				if (atl1c_get_link_status(hw)) {
 					if (atl1c_get_speed_and_duplex(hw, &speed,
 									&duplex) != 0)
 						dev_dbg(&pdev->dev,
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
index 73cbc049a63e..c263b326cec5 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_hw.h
@@ -26,6 +26,7 @@ void atl1c_phy_disable(struct atl1c_hw *hw);
 void atl1c_hw_set_mac_addr(struct atl1c_hw *hw, u8 *mac_addr);
 int atl1c_phy_reset(struct atl1c_hw *hw);
 int atl1c_read_mac_addr(struct atl1c_hw *hw);
+bool atl1c_get_link_status(struct atl1c_hw *hw);
 int atl1c_get_speed_and_duplex(struct atl1c_hw *hw, u16 *speed, u16 *duplex);
 u32 atl1c_hash_mc_addr(struct atl1c_hw *hw, u8 *mc_addr);
 void atl1c_hash_set(struct atl1c_hw *hw, u32 hash_value);
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index 9693da5028cf..740127a6a21d 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -232,15 +232,14 @@ static void atl1c_check_link_status(struct atl1c_adapter *adapter)
 	struct pci_dev    *pdev   = adapter->pdev;
 	int err;
 	unsigned long flags;
-	u16 speed, duplex, phy_data;
+	u16 speed, duplex;
+	bool link;
 
 	spin_lock_irqsave(&adapter->mdio_lock, flags);
-	/* MII_BMSR must read twise */
-	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
-	atl1c_read_phy_reg(hw, MII_BMSR, &phy_data);
+	link = atl1c_get_link_status(hw);
 	spin_unlock_irqrestore(&adapter->mdio_lock, flags);
 
-	if ((phy_data & BMSR_LSTATUS) == 0) {
+	if (!link) {
 		/* link down */
 		netif_carrier_off(netdev);
 		hw->hibernate = true;
@@ -284,16 +283,13 @@ static void atl1c_link_chg_event(struct atl1c_adapter *adapter)
 {
 	struct net_device *netdev = adapter->netdev;
 	struct pci_dev    *pdev   = adapter->pdev;
-	u16 phy_data;
-	u16 link_up;
+	bool link;
 
 	spin_lock(&adapter->mdio_lock);
-	atl1c_read_phy_reg(&adapter->hw, MII_BMSR, &phy_data);
-	atl1c_read_phy_reg(&adapter->hw, MII_BMSR, &phy_data);
+	link = atl1c_get_link_status(&adapter->hw);
 	spin_unlock(&adapter->mdio_lock);
-	link_up = phy_data & BMSR_LSTATUS;
 	/* notify upper layer link down ASAP */
-	if (!link_up) {
+	if (!link) {
 		if (netif_carrier_ok(netdev)) {
 			/* old link state: Up */
 			netif_carrier_off(netdev);
-- 
2.31.1

Re: [PATCH net-next v2 0/5] atl1c: support for Mikrotik 10/25G NIC features

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (refs/heads/master):

On Thu, 13 May 2021 14:43:21 +0300 you wrote:
> The new Mikrotik 10/25G NIC maintains compatibility with existing atl1c
> driver. However it does have new features.
> 
> This patch set adds support for reporting cards higher link speed, max-mtu,
> enables rx csum offload and improves tx performance.
> 
> v2:
>     - fixed xmit_more handling as pointed out by Eric Dumazet
>     - added a more reliable link detection on Mikrotik 10/25G NIC
>       since MDIO op emulation can occasionally fail
> 
> [...]

Here is the summary with links:
  - [net-next,v2,1/5] atl1c: show correct link speed on Mikrotik 10/25G NIC
    https://git.kernel.org/netdev/net-next/c/f19d4997fd1f
  - [net-next,v2,2/5] atl1c: improve performance by avoiding unnecessary pcie writes on xmit
    https://git.kernel.org/netdev/net-next/c/d7ab6419bdee
  - [net-next,v2,3/5] atl1c: adjust max mtu according to Mikrotik 10/25G NIC ability
    https://git.kernel.org/netdev/net-next/c/545fa3fb1e84
  - [net-next,v2,4/5] atl1c: enable rx csum offload on Mikrotik 10/25G NIC
    https://git.kernel.org/netdev/net-next/c/b0390009502b
  - [net-next,v2,5/5] atl1c: improve link detection reliability on Mikrotik 10/25G NIC
    https://git.kernel.org/netdev/net-next/c/ea0fbd05d7bd

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html