Re: [Patch v3 net-next 7/7] octeontx2-pf: ethtool physical link configuration

Hi Jakub,

-----Original Message-----
From: Jakub Kicinski <kuba@kernel.org>
Sent: Wednesday, February 3, 2021 6:59 AM
To: Hariprasad Kelam <redacted>
Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org;
davem@davemloft.net; willemdebruijn.kernel@gmail.com;
andrew@lunn.ch; Sunil Kovvuri Goutham [off-list ref]; Linu
Cherian [off-list ref]; Geethasowjanya Akula
[off-list ref]; Jerin Jacob Kollanukkaran [off-list ref];
Subbaraya Sundeep Bhatta [off-list ref]
Subject: [EXT] Re: [Patch v3 net-next 7/7] octeontx2-pf: ethtool physical link
configuration

On Sun, 31 Jan 2021 18:41:05 +0530 Hariprasad Kelam wrote:

quoted

From: Christina Jacob <redacted>

Register set_link_ksetting callback with driver such that link
configurations parameters like advertised mode,speed, duplex and
autoneg can be configured.

below command
ethtool -s eth0 advertise 0x1 speed 10 duplex full autoneg on

Signed-off-by: Christina Jacob <redacted>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: Hariprasad Kelam <redacted>
---
 .../ethernet/marvell/octeontx2/nic/otx2_ethtool.c  | 67
++++++++++++++++++++++
 1 file changed, 67 insertions(+)

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c

b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c
index d637815..74a62de 100644

--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c

@@ -1170,6 +1170,72 @@ static int otx2_get_link_ksettings(struct

net_device *netdev,

quoted

 	return 0;
 }

+static void otx2_get_advertised_mode(const struct ethtool_link_ksettings

*cmd,

quoted

+				     u64 *mode)
+{
+	u32 bit_pos;
+
+	/* Firmware does not support requesting multiple advertised modes
+	 * return first set bit
+	 */
+	bit_pos = find_first_bit(cmd->link_modes.advertising,
+				 __ETHTOOL_LINK_MODE_MASK_NBITS);
+	if (bit_pos != __ETHTOOL_LINK_MODE_MASK_NBITS)
+		*mode = bit_pos;
+}
+
+static int otx2_set_link_ksettings(struct net_device *netdev,
+				   const struct ethtool_link_ksettings *cmd) {
+	struct otx2_nic *pf = netdev_priv(netdev);
+	struct ethtool_link_ksettings req_ks;
+	struct ethtool_link_ksettings cur_ks;
+	struct cgx_set_link_mode_req *req;
+	struct mbox *mbox = &pf->mbox;
+	int err = 0;
+
+	/* save requested link settings */
+	memcpy(&req_ks, cmd, sizeof(struct ethtool_link_ksettings));

Why do you make this copy? The comment above does not help at all.

Agreed copy is not necessary . Added this copy for comparing advertised modes with supported modes.
Will fix this in next version.

quoted

+	memset(&cur_ks, 0, sizeof(struct ethtool_link_ksettings));
+
+	if (!ethtool_validate_speed(cmd->base.speed) ||
+	    !ethtool_validate_duplex(cmd->base.duplex))
+		return -EINVAL;
+
+	if (cmd->base.autoneg != AUTONEG_ENABLE &&
+	    cmd->base.autoneg != AUTONEG_DISABLE)
+		return -EINVAL;
+
+	otx2_get_link_ksettings(netdev, &cur_ks);
+
+	/* Check requested modes against supported modes by hardware */
+	if (!bitmap_subset(req_ks.link_modes.advertising,
+			   cur_ks.link_modes.supported,
+			   __ETHTOOL_LINK_MODE_MASK_NBITS))
+		return -EINVAL;
+
+	mutex_lock(&mbox->lock);
+	req = otx2_mbox_alloc_msg_cgx_set_link_mode(&pf->mbox);
+	if (!req) {
+		err = -ENOMEM;
+		goto end;
+	}
+
+	req->args.speed = req_ks.base.speed;
+	/* firmware expects 1 for half duplex and 0 for full duplex
+	 * hence inverting
+	 */
+	req->args.duplex = req_ks.base.duplex ^ 0x1;
+	req->args.an = req_ks.base.autoneg;
+	otx2_get_advertised_mode(&req_ks, &req->args.mode);

But that only returns the first bit set. What does the device actually do? What
if the user cleared a middle bit?

This is initial patch series to support advertised modes. Current firmware design is such that
It can handle only one advertised mode. Due to this limitation we are always checking
The first set bit in advertised modes and passing it to firmware.
Will add multi advertised mode support in near future.

Thanks,
Hariprasad k

quoted

+	err = otx2_sync_mbox_msg(&pf->mbox);
+end:
+	mutex_unlock(&mbox->lock);
+	return err;
+}
+
 static const struct ethtool_ops otx2_ethtool_ops = {
 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
 				     ETHTOOL_COALESCE_MAX_FRAMES, @@

-1200,6 +1266,7 @@ static

quoted

const struct ethtool_ops otx2_ethtool_ops = {
 	.get_fecparam		= otx2_get_fecparam,
 	.set_fecparam		= otx2_set_fecparam,
 	.get_link_ksettings     = otx2_get_link_ksettings,
+	.set_link_ksettings     = otx2_set_link_ksettings,
 };

 void otx2_set_ethtool_ops(struct net_device *netdev)

On Thu, 4 Feb 2021 17:37:41 +0000 Hariprasad Kelam wrote:

quoted

quoted

+	req->args.speed = req_ks.base.speed;
+	/* firmware expects 1 for half duplex and 0 for full duplex
+	 * hence inverting
+	 */
+	req->args.duplex = req_ks.base.duplex ^ 0x1;
+	req->args.an = req_ks.base.autoneg;
+	otx2_get_advertised_mode(&req_ks, &req->args.mode);  

But that only returns the first bit set. What does the device actually do? What
if the user cleared a middle bit?
  

This is initial patch series to support advertised modes. Current firmware design is such that
It can handle only one advertised mode. Due to this limitation we are always checking
The first set bit in advertised modes and passing it to firmware.
Will add multi advertised mode support in near future.

Looking at patch 6 it seems like the get side already supports multiple
modes, although the example output only lists supported no advertised.

Is the device actually doing IEEE autoneg or just configures the speed,
lanes etc. according to the link mode selected?