[PATCH net-next v2 1/3] Add flow control support flags to gianfar's capabilities

[PATCH net-next v2 1/3] Add flow control support flags to gianfar's capabilities

[Matei Pavaluca]

From: Pavaluca Matei-B46610 <matei.pavaluca@freescale.com>

The phy device supports 802.3x flow control, but the specific flags are not set
in the phy initialisation code. Flow control flags need to be added to the
supported capabilities of the phydev by the driver.

This is needed in order for ethtool to work ('ethtool -A' code checks for these
flags)

Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>
---

v2:
  - none
 
 drivers/net/ethernet/freescale/gianfar.c | 3 +++
 drivers/net/ethernet/freescale/gianfar.h | 4 +---
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
index 4fdf0aa..2485b74 100644
--- a/drivers/net/ethernet/freescale/gianfar.c
+++ b/drivers/net/ethernet/freescale/gianfar.c
@@ -1687,6 +1687,9 @@ static int init_phy(struct net_device *dev)
 	priv->phydev->supported &= (GFAR_SUPPORTED | gigabit_support);
 	priv->phydev->advertising = priv->phydev->supported;
 
+	/* Add support for flow control, but don't advertise it by default */
+	priv->phydev->supported |= (SUPPORTED_Pause | SUPPORTED_Asym_Pause);
+
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/freescale/gianfar.h b/drivers/net/ethernet/freescale/gianfar.h
index 2805cfb..6b00868 100644
--- a/drivers/net/ethernet/freescale/gianfar.h
+++ b/drivers/net/ethernet/freescale/gianfar.h
@@ -145,9 +145,7 @@ extern const char gfar_driver_version[];
 		| SUPPORTED_Autoneg \
 		| SUPPORTED_MII)
 
-#define GFAR_SUPPORTED_GBIT (SUPPORTED_1000baseT_Full \
-		| SUPPORTED_Pause \
-		| SUPPORTED_Asym_Pause)
+#define GFAR_SUPPORTED_GBIT SUPPORTED_1000baseT_Full
 
 /* TBI register addresses */
 #define MII_TBICON		0x11
-- 
1.7.11.7

[PATCH net-next v2 2/3] Fix the way the local advertising flow options are determined

[Matei Pavaluca]

From: Pavaluca Matei-B46610 <matei.pavaluca@freescale.com>

Local flow control options needed in order to resolve the negotiation
are incorrectly calculated.

Previously 'mii_advertise_flowctrl' was called to determine the local advertising
options, but these were determined based on FLOW_CTRL_RX/TX flags which are
never set through ethtool.
The patch simply translates from ethtool flow options to mii flow options.

Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>
---

 - none

 drivers/net/ethernet/freescale/gianfar.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
index 2485b74..329efca 100644
--- a/drivers/net/ethernet/freescale/gianfar.c
+++ b/drivers/net/ethernet/freescale/gianfar.c
@@ -3373,7 +3373,11 @@ static u32 gfar_get_flowctrl_cfg(struct gfar_private *priv)
 		if (phydev->asym_pause)
 			rmt_adv |= LPA_PAUSE_ASYM;
 
-		lcl_adv = mii_advertise_flowctrl(phydev->advertising);
+		lcl_adv = 0;
+		if (phydev->advertising & ADVERTISED_Pause)
+			lcl_adv |= ADVERTISE_PAUSE_CAP;
+		if (phydev->advertising & ADVERTISED_Asym_Pause)
+			lcl_adv |= ADVERTISE_PAUSE_ASYM;
 
 		flowctrl = mii_resolve_flowctrl_fdx(lcl_adv, rmt_adv);
 		if (flowctrl & FLOW_CTRL_TX)
-- 
1.7.11.7

Re: [PATCH net-next v2 2/3] Fix the way the local advertising flow options are determined

[David Miller]

From: Matei Pavaluca <matei.pavaluca@freescale.com>
Date: Mon, 27 Oct 2014 10:42:43 +0200

> From: Pavaluca Matei-B46610 <matei.pavaluca@freescale.com>
> 
> Local flow control options needed in order to resolve the negotiation
> are incorrectly calculated.
> 
> Previously 'mii_advertise_flowctrl' was called to determine the local advertising
> options, but these were determined based on FLOW_CTRL_RX/TX flags which are
> never set through ethtool.
> The patch simply translates from ethtool flow options to mii flow options.
> 
> Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>

Applied.

[PATCH net-next v2 3/3] gianfar: Implement PAUSE frame generation support

[Matei Pavaluca]

The hardware can automatically generate pause frames when the number
of free buffers drops under a certain threshold, but in order to do this,
the address of the last free buffer needs to be written to a specific
register for each RX queue.

This has to be done in 'gfar_clean_rx_ring' which is called for each
RX queue. In order not to impact performance, by adding a register write
for each incoming packet, this operation is done only when the PAUSE frame
transmission is enabled.

Whenever the link is readjusted, this capability is turned on or off.

Signed-off-by: Matei Pavaluca <matei.pavaluca@freescale.com>
---
 
v2:
  - Added unlikely() in 'gfar_clean_rx_ring'
  - Removed some unused variables
 
 drivers/net/ethernet/freescale/gianfar.c         | 54 ++++++++++++++++++++++++
 drivers/net/ethernet/freescale/gianfar.h         | 34 ++++++++++++++-
 drivers/net/ethernet/freescale/gianfar_ethtool.c |  7 ++-
 3 files changed, 93 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
index 329efca..86dccb26 100644
--- a/drivers/net/ethernet/freescale/gianfar.c
+++ b/drivers/net/ethernet/freescale/gianfar.c
@@ -173,10 +173,12 @@ static void gfar_init_rxbdp(struct gfar_priv_rx_q *rx_queue, struct rxbd8 *bdp,
 static int gfar_init_bds(struct net_device *ndev)
 {
 	struct gfar_private *priv = netdev_priv(ndev);
+	struct gfar __iomem *regs = priv->gfargrp[0].regs;
 	struct gfar_priv_tx_q *tx_queue = NULL;
 	struct gfar_priv_rx_q *rx_queue = NULL;
 	struct txbd8 *txbdp;
 	struct rxbd8 *rxbdp;
+	u32 *rfbptr;
 	int i, j;
 
 	for (i = 0; i < priv->num_tx_queues; i++) {
@@ -201,6 +203,7 @@ static int gfar_init_bds(struct net_device *ndev)
 		txbdp->status |= TXBD_WRAP;
 	}
 
+	rfbptr = &regs->rfbptr0;
 	for (i = 0; i < priv->num_rx_queues; i++) {
 		rx_queue = priv->rx_queue[i];
 		rx_queue->cur_rx = rx_queue->rx_bd_base;
@@ -227,6 +230,8 @@ static int gfar_init_bds(struct net_device *ndev)
 			rxbdp++;
 		}
 
+		rx_queue->rfbptr = rfbptr;
+		rfbptr += 2;
 	}
 
 	return 0;
@@ -336,6 +341,20 @@ static void gfar_init_tx_rx_base(struct gfar_private *priv)
 	}
 }
 
+static void gfar_init_rqprm(struct gfar_private *priv)
+{
+	struct gfar __iomem *regs = priv->gfargrp[0].regs;
+	u32 __iomem *baddr;
+	int i;
+
+	baddr = &regs->rqprm0;
+	for (i = 0; i < priv->num_rx_queues; i++) {
+		gfar_write(baddr, priv->rx_queue[i]->rx_ring_size |
+			   (DEFAULT_RX_LFC_THR << FBTHR_SHIFT));
+		baddr++;
+	}
+}
+
 static void gfar_rx_buff_size_config(struct gfar_private *priv)
 {
 	int frame_size = priv->ndev->mtu + ETH_HLEN + ETH_FCS_LEN;
@@ -396,6 +415,13 @@ static void gfar_mac_rx_config(struct gfar_private *priv)
 	if (priv->ndev->features & NETIF_F_HW_VLAN_CTAG_RX)
 		rctrl |= RCTRL_VLEX | RCTRL_PRSDEP_INIT;
 
+	/* Clear the LFC bit */
+	gfar_write(&regs->rctrl, rctrl);
+	/* Init flow control threshold values */
+	gfar_init_rqprm(priv);
+	gfar_write(&regs->ptv, DEFAULT_LFC_PTVVAL);
+	rctrl |= RCTRL_LFC;
+
 	/* Init rctrl based on our settings */
 	gfar_write(&regs->rctrl, rctrl);
 }
@@ -2859,6 +2885,10 @@ int gfar_clean_rx_ring(struct gfar_priv_rx_q *rx_queue, int rx_work_limit)
 		/* Setup the new bdp */
 		gfar_new_rxbdp(rx_queue, bdp, newskb);
 
+		/* Update Last Free RxBD pointer for LFC */
+		if (unlikely(rx_queue->rfbptr && priv->tx_actual_en))
+			gfar_write(rx_queue->rfbptr, (u32)bdp);
+
 		/* Update to the next pointer */
 		bdp = next_bd(bdp, base, rx_queue->rx_ring_size);
 
@@ -3393,6 +3423,9 @@ static noinline void gfar_update_link_state(struct gfar_private *priv)
 {
 	struct gfar __iomem *regs = priv->gfargrp[0].regs;
 	struct phy_device *phydev = priv->phydev;
+	struct gfar_priv_rx_q *rx_queue = NULL;
+	int i;
+	struct rxbd8 *bdp;
 
 	if (unlikely(test_bit(GFAR_RESETTING, &priv->state)))
 		return;
@@ -3401,6 +3434,7 @@ static noinline void gfar_update_link_state(struct gfar_private *priv)
 		u32 tempval1 = gfar_read(&regs->maccfg1);
 		u32 tempval = gfar_read(&regs->maccfg2);
 		u32 ecntrl = gfar_read(&regs->ecntrl);
+		u32 tx_flow_oldval = (tempval & MACCFG1_TX_FLOW);
 
 		if (phydev->duplex != priv->oldduplex) {
 			if (!(phydev->duplex))
@@ -3445,6 +3479,26 @@ static noinline void gfar_update_link_state(struct gfar_private *priv)
 		tempval1 &= ~(MACCFG1_TX_FLOW | MACCFG1_RX_FLOW);
 		tempval1 |= gfar_get_flowctrl_cfg(priv);
 
+		/* Turn last free buffer recording on */
+		if ((tempval1 & MACCFG1_TX_FLOW) && !tx_flow_oldval) {
+			for (i = 0; i < priv->num_rx_queues; i++) {
+				rx_queue = priv->rx_queue[i];
+				bdp = rx_queue->cur_rx;
+				/* skip to previous bd */
+				bdp = skip_bd(bdp, rx_queue->rx_ring_size - 1,
+					      rx_queue->rx_bd_base,
+					      rx_queue->rx_ring_size);
+
+				if (rx_queue->rfbptr)
+					gfar_write(rx_queue->rfbptr, (u32)bdp);
+			}
+
+			priv->tx_actual_en = 1;
+		}
+
+		if (unlikely(!(tempval1 & MACCFG1_TX_FLOW) && tx_flow_oldval))
+			priv->tx_actual_en = 0;
+
 		gfar_write(&regs->maccfg1, tempval1);
 		gfar_write(&regs->maccfg2, tempval);
 		gfar_write(&regs->ecntrl, ecntrl);
diff --git a/drivers/net/ethernet/freescale/gianfar.h b/drivers/net/ethernet/freescale/gianfar.h
index 6b00868..b581b88 100644
--- a/drivers/net/ethernet/freescale/gianfar.h
+++ b/drivers/net/ethernet/freescale/gianfar.h
@@ -99,6 +99,10 @@ extern const char gfar_driver_version[];
 #define GFAR_MAX_FIFO_STARVE	511
 #define GFAR_MAX_FIFO_STARVE_OFF 511
 
+#define FBTHR_SHIFT        24
+#define DEFAULT_RX_LFC_THR  16
+#define DEFAULT_LFC_PTVVAL  4
+
 #define DEFAULT_RX_BUFFER_SIZE  1536
 #define TX_RING_MOD_MASK(size) (size-1)
 #define RX_RING_MOD_MASK(size) (size-1)
@@ -273,6 +277,7 @@ extern const char gfar_driver_version[];
 
 #define RCTRL_TS_ENABLE 	0x01000000
 #define RCTRL_PAL_MASK		0x001f0000
+#define RCTRL_LFC		0x00004000
 #define RCTRL_VLEX		0x00002000
 #define RCTRL_FILREN		0x00001000
 #define RCTRL_GHTX		0x00000400
@@ -849,7 +854,32 @@ struct gfar {
 	u8	res23c[248];
 	u32	attr;		/* 0x.bf8 - Attributes Register */
 	u32	attreli;	/* 0x.bfc - Attributes Extract Length and Extract Index Register */
-	u8	res24[688];
+	u32	rqprm0;	/* 0x.c00 - Receive queue parameters register 0 */
+	u32	rqprm1;	/* 0x.c04 - Receive queue parameters register 1 */
+	u32	rqprm2;	/* 0x.c08 - Receive queue parameters register 2 */
+	u32	rqprm3;	/* 0x.c0c - Receive queue parameters register 3 */
+	u32	rqprm4;	/* 0x.c10 - Receive queue parameters register 4 */
+	u32	rqprm5;	/* 0x.c14 - Receive queue parameters register 5 */
+	u32	rqprm6;	/* 0x.c18 - Receive queue parameters register 6 */
+	u32	rqprm7;	/* 0x.c1c - Receive queue parameters register 7 */
+	u8	res24[36];
+	u32	rfbptr0; /* 0x.c44 - Last free RxBD pointer for ring 0 */
+	u8	res24a[4];
+	u32	rfbptr1; /* 0x.c4c - Last free RxBD pointer for ring 1 */
+	u8	res24b[4];
+	u32	rfbptr2; /* 0x.c54 - Last free RxBD pointer for ring 2 */
+	u8	res24c[4];
+	u32	rfbptr3; /* 0x.c5c - Last free RxBD pointer for ring 3 */
+	u8	res24d[4];
+	u32	rfbptr4; /* 0x.c64 - Last free RxBD pointer for ring 4 */
+	u8	res24e[4];
+	u32	rfbptr5; /* 0x.c6c - Last free RxBD pointer for ring 5 */
+	u8	res24f[4];
+	u32	rfbptr6; /* 0x.c74 - Last free RxBD pointer for ring 6 */
+	u8	res24g[4];
+	u32	rfbptr7; /* 0x.c7c - Last free RxBD pointer for ring 7 */
+	u8	res24h[4];
+	u8	res24x[556];
 	u32	isrg0;		/* 0x.eb0 - Interrupt steering group 0 register */
 	u32	isrg1;		/* 0x.eb4 - Interrupt steering group 1 register */
 	u32	isrg2;		/* 0x.eb8 - Interrupt steering group 2 register */
@@ -1009,6 +1039,7 @@ struct gfar_priv_rx_q {
 	/* RX Coalescing values */
 	unsigned char rxcoalescing;
 	unsigned long rxic;
+	u32 *rfbptr;
 };
 
 enum gfar_irqinfo_id {
@@ -1099,6 +1130,7 @@ struct gfar_private {
 	unsigned int num_tx_queues;
 	unsigned int num_rx_queues;
 	unsigned int num_grps;
+	int tx_actual_en;
 
 	/* Network Statistics */
 	struct gfar_extra_stats extra_stats;
diff --git a/drivers/net/ethernet/freescale/gianfar_ethtool.c b/drivers/net/ethernet/freescale/gianfar_ethtool.c
index 76d7070..3e1a9c1 100644
--- a/drivers/net/ethernet/freescale/gianfar_ethtool.c
+++ b/drivers/net/ethernet/freescale/gianfar_ethtool.c
@@ -579,8 +579,13 @@ static int gfar_spauseparam(struct net_device *dev,
 			u32 tempval;
 			tempval = gfar_read(&regs->maccfg1);
 			tempval &= ~(MACCFG1_TX_FLOW | MACCFG1_RX_FLOW);
-			if (priv->tx_pause_en)
+
+			priv->tx_actual_en = 0;
+			if (priv->tx_pause_en) {
+				priv->tx_actual_en = 1;
 				tempval |= MACCFG1_TX_FLOW;
+			}
+
 			if (priv->rx_pause_en)
 				tempval |= MACCFG1_RX_FLOW;
 			gfar_write(&regs->maccfg1, tempval);
-- 
1.7.11.7

Re: [PATCH net-next v2 3/3] gianfar: Implement PAUSE frame generation support

[David Miller]

From: Matei Pavaluca <matei.pavaluca@freescale.com>
Date: Mon, 27 Oct 2014 10:42:44 +0200

> The hardware can automatically generate pause frames when the number
> of free buffers drops under a certain threshold, but in order to do this,
> the address of the last free buffer needs to be written to a specific
> register for each RX queue.
> 
> This has to be done in 'gfar_clean_rx_ring' which is called for each
> RX queue. In order not to impact performance, by adding a register write
> for each incoming packet, this operation is done only when the PAUSE frame
> transmission is enabled.
> 
> Whenever the link is readjusted, this capability is turned on or off.
> 
> Signed-off-by: Matei Pavaluca <matei.pavaluca@freescale.com>

Applied.

But I would turn pause frames on by default, and deal with the
"performance" issue by aggregating the register writes.  Only
do it every N cleaned frames.

Re: [PATCH net-next v2 1/3] Add flow control support flags to gianfar's capabilities

[David Miller]

From: Matei Pavaluca <matei.pavaluca@freescale.com>
Date: Mon, 27 Oct 2014 10:42:42 +0200

> From: Pavaluca Matei-B46610 <matei.pavaluca@freescale.com>
> 
> The phy device supports 802.3x flow control, but the specific flags are not set
> in the phy initialisation code. Flow control flags need to be added to the
> supported capabilities of the phydev by the driver.
> 
> This is needed in order for ethtool to work ('ethtool -A' code checks for these
> flags)
> 
> Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>

Applied.