Hi Sergei,

Subject: Re: [PATCH net-next 12/18] ravb: Factorise {emac,dmac} init
function

On 8/20/21 6:42 PM, Biju Das wrote:

[...]

quoted

quoted

quoted

The R-Car AVB module has Magic packet detection, multiple irq's and
timestamp enable features which is not present on RZ/G2L Gigabit

                                   ^ are

OK. Will fix this in next patch set.

quoted

quoted

Ethernet module. Factorise emac and dmac initialization function to
support the later SoC.

Signed-off-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
---
 drivers/net/ethernet/renesas/ravb.h      |  2 +
 drivers/net/ethernet/renesas/ravb_main.c | 58
++++++++++++++++--------
 2 files changed, 40 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ethernet/renesas/ravb.h

b/drivers/net/ethernet/renesas/ravb.h
index d82bfa6e57c1..4d5910dcda86 100644

--- a/drivers/net/ethernet/renesas/ravb.h
+++ b/drivers/net/ethernet/renesas/ravb.h

@@ -992,6 +992,8 @@ struct ravb_ops {
 	void (*ring_free)(struct net_device *ndev, int q);
 	void (*ring_format)(struct net_device *ndev, int q);
 	bool (*alloc_rx_desc)(struct net_device *ndev, int q);
+	void (*emac_init)(struct net_device *ndev);
+	void (*dmac_init)(struct net_device *ndev);
 };

 struct ravb_drv_data {

diff --git a/drivers/net/ethernet/renesas/ravb_main.c

b/drivers/net/ethernet/renesas/ravb_main.c
index 3d0f6598b936..e200114376e4 100644

--- a/drivers/net/ethernet/renesas/ravb_main.c
+++ b/drivers/net/ethernet/renesas/ravb_main.c

@@ -454,7 +454,7 @@ static int ravb_ring_init(struct net_device

*ndev, int q)  }

 /* E-MAC init function */
-static void ravb_emac_init(struct net_device *ndev)
+static void ravb_emac_init_ex(struct net_device *ndev)
 {
 	/* Receive frame limit set register */
 	ravb_write(ndev, ndev->mtu + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN,
RFLR); @@ -480,30 +480,19 @@ static void ravb_emac_init(struct

net_device *ndev)

quoted

 	ravb_write(ndev, ECSIPR_ICDIP | ECSIPR_MPDIP | ECSIPR_LCHNGIP,
ECSIPR);  }

-/* Device init function for Ethernet AVB */

   Grr, this comment seems oudated...

OK.

   Just don't move the comment. :-)

quoted

quoted

quoted

-static int ravb_dmac_init(struct net_device *ndev)
+static void ravb_emac_init(struct net_device *ndev)
 {
 	struct ravb_private *priv = netdev_priv(ndev);
 	const struct ravb_drv_data *info = priv->info;
-	int error;

-	/* Set CONFIG mode */
-	error = ravb_config(ndev);
-	if (error)
-		return error;
-
-	error = ravb_ring_init(ndev, RAVB_BE);
-	if (error)
-		return error;
-	error = ravb_ring_init(ndev, RAVB_NC);
-	if (error) {
-		ravb_ring_free(ndev, RAVB_BE);
-		return error;
-	}
+	info->ravb_ops->emac_init(ndev);
+}

   The whole ravb_emac_init() now consists only of a single method

call?

quoted

quoted

Why do we need it at all?

OK will assign info->emac_init with ravb_emac_init, so GbEthernet just
need to fill emac_init function. I will remove the function

"ravb_emac_init_ex".

  Will the EMAC init methods differ so much as to we should provide 2
separate implementations?

[...]

quoted

quoted

quoted

+static void ravb_dmac_init_ex(struct net_device *ndev)

   Please no _ex suffixes -- reminds me of Windoze too much. :-)

OK. Will change it to ravb_device_init

   Ugh! Why not leave it named ravb_dmac_init()?

Please see [1] below and also planning to send another 10 small incremental patches,
So that it is clear to every one.

quoted

Regards,
Biju

quoted

quoted

+{
+	struct ravb_private *priv = netdev_priv(ndev);
+	const struct ravb_drv_data *info = priv->info;

 	/* Set AVB RX */
 	ravb_write(ndev,

@@ -530,6 +519,33 @@ static int ravb_dmac_init(struct net_device

*ndev)

quoted

quoted

quoted

 	ravb_write(ndev, RIC2_QFE0 | RIC2_QFE1 | RIC2_RFFE, RIC2);
 	/* Frame transmitted, timestamp FIFO updated */
 	ravb_write(ndev, TIC_FTE0 | TIC_FTE1 | TIC_TFUE, TIC);
+}
+
+static int ravb_dmac_init(struct net_device *ndev) {
+	struct ravb_private *priv = netdev_priv(ndev);
+	const struct ravb_drv_data *info = priv->info;
+	int error;
+
+	/* Set CONFIG mode */
+	error = ravb_config(ndev);
+	if (error)
+		return error;
+
+	error = ravb_ring_init(ndev, RAVB_BE);
+	if (error)
+		return error;
+	error = ravb_ring_init(ndev, RAVB_NC);
+	if (error) {
+		ravb_ring_free(ndev, RAVB_BE);
+		return error;
+	}
+
+	/* Descriptor format */
+	ravb_ring_format(ndev, RAVB_BE);
+	ravb_ring_format(ndev, RAVB_NC);
+
+	info->ravb_ops->dmac_init(ndev);

 	/* Setting the control will start the AVB-DMAC process. */
 	ravb_modify(ndev, CCC, CCC_OPC, CCC_OPC_OPERATION); @@ -2018,6
+2034,8 @@ static const struct ravb_ops ravb_gen3_ops = {
 	.ring_free = ravb_ring_free_rx,
 	.ring_format = ravb_ring_format_rx,
 	.alloc_rx_desc = ravb_alloc_rx_desc,
+	.emac_init = ravb_emac_init_ex,
+	.dmac_init = ravb_dmac_init_ex,

   Hmm, why not also gen2?!

   The question remained unreplied?... :-/

ravb_ops for gen3 and gen2 same. But RZ/G2L have different function pointers[1].

As agreed on next patchset, we are avoiding indirection and it will be hw_ifo.
In that case there will be 1 entry in gen3 and 1 entry in gen2. Hope this clears
Your doubts.

[1] https://patchwork.kernel.org/project/linux-renesas-soc/patch/20210722141351.13668-18-biju.das.jz@bp.renesas.com/

Cheers,
Biju