diff mbox series

[net-next,v2,10/10] bnxt_en: Implement .ndo_features_check().

Message ID 1619372727-19187-11-git-send-email-michael.chan@broadcom.com
State New
Headers show
Series bnxt_en: Updates for net-next. | expand

Commit Message

Michael Chan April 25, 2021, 5:45 p.m. UTC 
For UDP encapsultions, we only support the offloaded Vxlan port and
Geneve port.  All other ports included FOU and GUE are not supported so
we need to turn off TSO and checksum features.

v2: Reverse the check for supported UDP ports to be more straight forward.

Reviewed-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 42 +++++++++++++++++++++--
 drivers/net/ethernet/broadcom/bnxt/bnxt.h |  2 ++
 2 files changed, 42 insertions(+), 2 deletions(-)

Comments

Jakub Kicinski April 26, 2021, 4:29 p.m. UTC | #1 
On Sun, 25 Apr 2021 13:45:27 -0400 Michael Chan wrote:
> +	if (l4_proto != IPPROTO_UDP)

> +		return features;


This should have been "features & ~(CSUM | GSO)" if you actually
accepted my feedback. I mentioned extension headers as an example, 
v2 looks like a minor refactoring, no functional changes :/

> +	bp = netdev_priv(dev);

> +	/* For UDP, we can only handle 1 Vxlan port and 1 Geneve port. */

> +	udp_port = udp_hdr(skb)->dest;

> +	if (udp_port == bp->vxlan_port || udp_port == bp->nge_port)

> +		return features;

> +	return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
Michael Chan April 26, 2021, 4:35 p.m. UTC | #2 
On Mon, Apr 26, 2021 at 9:29 AM Jakub Kicinski <kuba@kernel.org> wrote:
>

> On Sun, 25 Apr 2021 13:45:27 -0400 Michael Chan wrote:

> > +     if (l4_proto != IPPROTO_UDP)

> > +             return features;

>

> This should have been "features & ~(CSUM | GSO)" if you actually

> accepted my feedback. I mentioned extension headers as an example,

> v2 looks like a minor refactoring, no functional changes :/

>

> > +     bp = netdev_priv(dev);

> > +     /* For UDP, we can only handle 1 Vxlan port and 1 Geneve port. */

> > +     udp_port = udp_hdr(skb)->dest;

> > +     if (udp_port == bp->vxlan_port || udp_port == bp->nge_port)

> > +             return features;

> > +     return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
Michael Chan April 26, 2021, 4:45 p.m. UTC | #3 
On Mon, Apr 26, 2021 at 9:35 AM Michael Chan <michael.chan@broadcom.com> wrote:
>

> On Mon, Apr 26, 2021 at 9:29 AM Jakub Kicinski <kuba@kernel.org> wrote:

> >

> > On Sun, 25 Apr 2021 13:45:27 -0400 Michael Chan wrote:

> > > +     if (l4_proto != IPPROTO_UDP)

> > > +             return features;

> >

> > This should have been "features & ~(CSUM | GSO)" if you actually

> > accepted my feedback.


Sorry, hit send too early.  If it is not UDP, it could be GRE or IP
over IP for example, right?  Why do we need to turn off offload?

> I mentioned extension headers as an example,


Extension headers (Geneve for example) are supported.
Jakub Kicinski April 26, 2021, 5 p.m. UTC | #4 
On Mon, 26 Apr 2021 09:45:17 -0700 Michael Chan wrote:
> On Mon, Apr 26, 2021 at 9:35 AM Michael Chan <michael.chan@broadcom.com> wrote:

> > On Mon, Apr 26, 2021 at 9:29 AM Jakub Kicinski <kuba@kernel.org> wrote:  

> > > On Sun, 25 Apr 2021 13:45:27 -0400 Michael Chan wrote:  

> > > This should have been "features & ~(CSUM | GSO)" if you actually

> > > accepted my feedback.  

> 

> Sorry, hit send too early.  If it is not UDP, it could be GRE or IP

> over IP for example, right?  Why do we need to turn off offload?


All supported protocols can be included in the allow list.
That's one of the costs of NETIF_F_IP_CSUM, the driver needs 
to make sure the device can understand the packet.

> > I mentioned extension headers as an example,  

> 

> Extension headers (Geneve for example) are supported.


I thought the Geneve things were called options. I meant IPv6 extension
headers, which the device may also support, but then the right thing to
do is something like a call to ipv6_skip_exthdr() to retrieve the L4
proto.
Michael Chan April 26, 2021, 5:09 p.m. UTC | #5 
On Mon, Apr 26, 2021 at 10:00 AM Jakub Kicinski <kuba@kernel.org> wrote:
>

> On Mon, 26 Apr 2021 09:45:17 -0700 Michael Chan wrote:

> > On Mon, Apr 26, 2021 at 9:35 AM Michael Chan <michael.chan@broadcom.com> wrote:

> > > On Mon, Apr 26, 2021 at 9:29 AM Jakub Kicinski <kuba@kernel.org> wrote:

> > > > On Sun, 25 Apr 2021 13:45:27 -0400 Michael Chan wrote:

> > > > This should have been "features & ~(CSUM | GSO)" if you actually

> > > > accepted my feedback.

> >

> > Sorry, hit send too early.  If it is not UDP, it could be GRE or IP

> > over IP for example, right?  Why do we need to turn off offload?

>

> All supported protocols can be included in the allow list.

> That's one of the costs of NETIF_F_IP_CSUM, the driver needs

> to make sure the device can understand the packet.


Only UDP encapsulations have the 2 port limitations.  The rest are supported.

>

> > > I mentioned extension headers as an example,

> >

> > Extension headers (Geneve for example) are supported.

>

> I thought the Geneve things were called options. I meant IPv6 extension

> headers, which the device may also support, but then the right thing to

> do is something like a call to ipv6_skip_exthdr() to retrieve the L4

> proto.


You're right and I misunderstood you.  I will double check for ipv6
extension headers support and will send a follow up patch.  Thanks.
Jakub Kicinski April 26, 2021, 5:36 p.m. UTC | #6 
On Mon, 26 Apr 2021 10:09:29 -0700 Michael Chan wrote:
> On Mon, Apr 26, 2021 at 10:00 AM Jakub Kicinski <kuba@kernel.org> wrote:

> > On Mon, 26 Apr 2021 09:45:17 -0700 Michael Chan wrote:  

> > > Sorry, hit send too early.  If it is not UDP, it could be GRE or IP

> > > over IP for example, right?  Why do we need to turn off offload?  

> >

> > All supported protocols can be included in the allow list.

> > That's one of the costs of NETIF_F_IP_CSUM, the driver needs

> > to make sure the device can understand the packet.  

> 

> Only UDP encapsulations have the 2 port limitations.  The rest are supported.


AFAIU you claim that other than certain UDP formats your device can
parse _every_ possible packet encapsulation. To which I have no reply.
Michael Chan April 26, 2021, 5:52 p.m. UTC | #7 
On Mon, Apr 26, 2021 at 10:36 AM Jakub Kicinski <kuba@kernel.org> wrote:
>

> On Mon, 26 Apr 2021 10:09:29 -0700 Michael Chan wrote:

> > On Mon, Apr 26, 2021 at 10:00 AM Jakub Kicinski <kuba@kernel.org> wrote:

> > > On Mon, 26 Apr 2021 09:45:17 -0700 Michael Chan wrote:

> > > > Sorry, hit send too early.  If it is not UDP, it could be GRE or IP

> > > > over IP for example, right?  Why do we need to turn off offload?

> > >

> > > All supported protocols can be included in the allow list.

> > > That's one of the costs of NETIF_F_IP_CSUM, the driver needs

> > > to make sure the device can understand the packet.

> >

> > Only UDP encapsulations have the 2 port limitations.  The rest are supported.

>

> AFAIU you claim that other than certain UDP formats your device can

> parse _every_ possible packet encapsulation. To which I have no reply.


Fair enough.  I'll get the list of supported encapsulations and check
for all supported types.
diff mbox series

Patch

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 53db073b457c..5d0ab5629aa4 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -10781,6 +10781,40 @@  static int bnxt_set_features(struct net_device *dev, netdev_features_t features)
 	return rc;
 }
 
+static netdev_features_t bnxt_features_check(struct sk_buff *skb,
+					     struct net_device *dev,
+					     netdev_features_t features)
+{
+	struct bnxt *bp;
+	__be16 udp_port;
+	u8 l4_proto = 0;
+
+	features = vlan_features_check(skb, features);
+	if (!skb->encapsulation)
+		return features;
+
+	switch (vlan_get_protocol(skb)) {
+	case htons(ETH_P_IP):
+		l4_proto = ip_hdr(skb)->protocol;
+		break;
+	case htons(ETH_P_IPV6):
+		l4_proto = ipv6_hdr(skb)->nexthdr;
+		break;
+	default:
+		return features;
+	}
+
+	if (l4_proto != IPPROTO_UDP)
+		return features;
+
+	bp = netdev_priv(dev);
+	/* For UDP, we can only handle 1 Vxlan port and 1 Geneve port. */
+	udp_port = udp_hdr(skb)->dest;
+	if (udp_port == bp->vxlan_port || udp_port == bp->nge_port)
+		return features;
+	return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
+}
+
 int bnxt_dbg_hwrm_rd_reg(struct bnxt *bp, u32 reg_off, u16 num_words,
 			 u32 *reg_buf)
 {
@@ -12288,10 +12322,13 @@  static int bnxt_udp_tunnel_sync(struct net_device *netdev, unsigned int table)
 	unsigned int cmd;
 
 	udp_tunnel_nic_get_port(netdev, table, 0, &ti);
-	if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
+	if (ti.type == UDP_TUNNEL_TYPE_VXLAN) {
+		bp->vxlan_port = ti.port;
 		cmd = TUNNEL_DST_PORT_FREE_REQ_TUNNEL_TYPE_VXLAN;
-	else
+	} else {
+		bp->nge_port = ti.port;
 		cmd = TUNNEL_DST_PORT_FREE_REQ_TUNNEL_TYPE_GENEVE;
+	}
 
 	if (ti.port)
 		return bnxt_hwrm_tunnel_dst_port_alloc(bp, ti.port, cmd);
@@ -12391,6 +12428,7 @@  static const struct net_device_ops bnxt_netdev_ops = {
 	.ndo_change_mtu		= bnxt_change_mtu,
 	.ndo_fix_features	= bnxt_fix_features,
 	.ndo_set_features	= bnxt_set_features,
+	.ndo_features_check	= bnxt_features_check,
 	.ndo_tx_timeout		= bnxt_tx_timeout,
 #ifdef CONFIG_BNXT_SRIOV
 	.ndo_get_vf_config	= bnxt_get_vf_config,
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index a3744247740b..24d2ad6a8740 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -1914,6 +1914,8 @@  struct bnxt {
 
 	u16			vxlan_fw_dst_port_id;
 	u16			nge_fw_dst_port_id;
+	__be16			vxlan_port;
+	__be16			nge_port;
 	u8			port_partition_type;
 	u8			port_count;
 	u16			br_mode;