diff mbox

[net-next,4/7] devlink: Add E-Switch inline mode control

Message ID 1479733561-26601-5-git-send-email-saeedm@mellanox.com
State Changes Requested, archived
Delegated to: David Miller
Headers show

Commit Message

Saeed Mahameed Nov. 21, 2016, 1:05 p.m. UTC 
From: Roi Dayan <roid@mellanox.com>

Some HWs need the VF driver to put part of the packet headers on the
TX descriptor so the e-switch can do proper matching and steering.

The supported modes: none, link, network, transport.

Signed-off-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 include/net/devlink.h        |  2 ++
 include/uapi/linux/devlink.h |  8 +++++
 net/core/devlink.c           | 70 ++++++++++++++++++++++++++++++++------------
 3 files changed, 61 insertions(+), 19 deletions(-)

Comments

Jiri Pirko Jan. 8, 2017, 10:29 a.m. UTC | #1 
Mon, Nov 21, 2016 at 02:05:58PM CET, saeedm@mellanox.com wrote:
>From: Roi Dayan <roid@mellanox.com>
>
>Some HWs need the VF driver to put part of the packet headers on the
>TX descriptor so the e-switch can do proper matching and steering.

Could you please elaborate a bit about possible use-cases for different
modes? Thanks.


>
>The supported modes: none, link, network, transport.
>
>Signed-off-by: Roi Dayan <roid@mellanox.com>
>Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Or Gerlitz Jan. 8, 2017, 10:49 a.m. UTC | #2 
On 1/8/2017 12:29 PM, Jiri Pirko wrote:
> Mon, Nov 21, 2016 at 02:05:58PM CET, saeedm@mellanox.com wrote:
>> From: Roi Dayan <roid@mellanox.com>
>>
>> Some HWs need the VF driver to put part of the packet headers on the
>> TX descriptor so the e-switch can do proper matching and steering.
> Could you please elaborate a bit about possible use-cases for different
> modes? Thanks.

As written in the change log, some HW models have this requirement that 
the header set you want eswitch matching on (e.g L2/L3) is present as MD 
on the xmit DMA descriptor.

To address these requirements, following the admin devlink directive the 
FW advertizes that
to the VF, they are doing so in their xmit logic and the host driver 
enforces that the VF has the proper inline mode before we are willing to 
offload eswitch matching rules. If the VF doesn't obey to the 
requirement the packets are dropped by HW.

Or.

Or.
Jiri Pirko Jan. 8, 2017, 10:54 a.m. UTC | #3 
Sun, Jan 08, 2017 at 11:49:20AM CET, ogerlitz@mellanox.com wrote:
>On 1/8/2017 12:29 PM, Jiri Pirko wrote:
>> Mon, Nov 21, 2016 at 02:05:58PM CET, saeedm@mellanox.com wrote:
>> > From: Roi Dayan <roid@mellanox.com>
>> > 
>> > Some HWs need the VF driver to put part of the packet headers on the
>> > TX descriptor so the e-switch can do proper matching and steering.
>> Could you please elaborate a bit about possible use-cases for different
>> modes? Thanks.
>
>As written in the change log, some HW models have this requirement that the
>header set you want eswitch matching on (e.g L2/L3) is present as MD on the
>xmit DMA descriptor.
>
>To address these requirements, following the admin devlink directive the FW
>advertizes that
>to the VF, they are doing so in their xmit logic and the host driver enforces
>that the VF has the proper inline mode before we are willing to offload
>eswitch matching rules. If the VF doesn't obey to the requirement the packets
>are dropped by HW.

Okay, makes sense. Do you expect this will ever have to be needed
per-vf? In general, not only for mlx* drivers. I believe that this is
an e-switch requirement so it should be same for all connected VFs,
right?

Thanks!
Or Gerlitz Jan. 8, 2017, 11:10 a.m. UTC | #4 
On 1/8/2017 12:54 PM, Jiri Pirko wrote:
> I believe that this is an e-switch requirement so it should be same for all connected VFs, right?
yes
diff mbox

Patch

diff --git a/include/net/devlink.h b/include/net/devlink.h
index 211bd3c..d29e5fc 100644
--- a/include/net/devlink.h
+++ b/include/net/devlink.h
@@ -92,6 +92,8 @@  struct devlink_ops {
 
 	int (*eswitch_mode_get)(struct devlink *devlink, u16 *p_mode);
 	int (*eswitch_mode_set)(struct devlink *devlink, u16 mode);
+	int (*eswitch_inline_mode_get)(struct devlink *devlink, u8 *p_inline_mode);
+	int (*eswitch_inline_mode_set)(struct devlink *devlink, u8 inline_mode);
 };
 
 static inline void *devlink_priv(struct devlink *devlink)
diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h
index 915bfa7..9014c33 100644
--- a/include/uapi/linux/devlink.h
+++ b/include/uapi/linux/devlink.h
@@ -102,6 +102,13 @@  enum devlink_eswitch_mode {
 	DEVLINK_ESWITCH_MODE_SWITCHDEV,
 };
 
+enum devlink_eswitch_inline_mode {
+	DEVLINK_ESWITCH_INLINE_MODE_NONE,
+	DEVLINK_ESWITCH_INLINE_MODE_LINK,
+	DEVLINK_ESWITCH_INLINE_MODE_NETWORK,
+	DEVLINK_ESWITCH_INLINE_MODE_TRANSPORT,
+};
+
 enum devlink_attr {
 	/* don't change the order or add anything between, this is ABI! */
 	DEVLINK_ATTR_UNSPEC,
@@ -133,6 +140,7 @@  enum devlink_attr {
 	DEVLINK_ATTR_SB_OCC_CUR,		/* u32 */
 	DEVLINK_ATTR_SB_OCC_MAX,		/* u32 */
 	DEVLINK_ATTR_ESWITCH_MODE,		/* u16 */
+	DEVLINK_ATTR_ESWITCH_INLINE_MODE,	/* u8 */
 
 	/* add new attributes above here, update the policy in devlink.c */
 
diff --git a/net/core/devlink.c b/net/core/devlink.c
index c14f8b6..2b5bf9e 100644
--- a/net/core/devlink.c
+++ b/net/core/devlink.c
@@ -1394,26 +1394,45 @@  static int devlink_nl_cmd_sb_occ_max_clear_doit(struct sk_buff *skb,
 
 static int devlink_eswitch_fill(struct sk_buff *msg, struct devlink *devlink,
 				enum devlink_command cmd, u32 portid,
-				u32 seq, int flags, u16 mode)
+				u32 seq, int flags)
 {
+	const struct devlink_ops *ops = devlink->ops;
 	void *hdr;
+	int err = 0;
+	u16 mode;
+	u8 inline_mode;
 
 	hdr = genlmsg_put(msg, portid, seq, &devlink_nl_family, flags, cmd);
 	if (!hdr)
 		return -EMSGSIZE;
 
-	if (devlink_nl_put_handle(msg, devlink))
-		goto nla_put_failure;
+	err = devlink_nl_put_handle(msg, devlink);
+	if (err)
+		goto out;
 
-	if (nla_put_u16(msg, DEVLINK_ATTR_ESWITCH_MODE, mode))
-		goto nla_put_failure;
+	err = ops->eswitch_mode_get(devlink, &mode);
+	if (err)
+		goto out;
+	err = nla_put_u16(msg, DEVLINK_ATTR_ESWITCH_MODE, mode);
+	if (err)
+		goto out;
+
+	if (ops->eswitch_inline_mode_get) {
+		err = ops->eswitch_inline_mode_get(devlink, &inline_mode);
+		if (err)
+			goto out;
+		err = nla_put_u8(msg, DEVLINK_ATTR_ESWITCH_INLINE_MODE,
+				 inline_mode);
+		if (err)
+			goto out;
+	}
 
 	genlmsg_end(msg, hdr);
 	return 0;
 
-nla_put_failure:
+out:
 	genlmsg_cancel(msg, hdr);
-	return -EMSGSIZE;
+	return err;
 }
 
 static int devlink_nl_cmd_eswitch_mode_get_doit(struct sk_buff *skb,
@@ -1422,22 +1441,17 @@  static int devlink_nl_cmd_eswitch_mode_get_doit(struct sk_buff *skb,
 	struct devlink *devlink = info->user_ptr[0];
 	const struct devlink_ops *ops = devlink->ops;
 	struct sk_buff *msg;
-	u16 mode;
 	int err;
 
 	if (!ops || !ops->eswitch_mode_get)
 		return -EOPNOTSUPP;
 
-	err = ops->eswitch_mode_get(devlink, &mode);
-	if (err)
-		return err;
-
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!msg)
 		return -ENOMEM;
 
 	err = devlink_eswitch_fill(msg, devlink, DEVLINK_CMD_ESWITCH_MODE_GET,
-				   info->snd_portid, info->snd_seq, 0, mode);
+				   info->snd_portid, info->snd_seq, 0);
 
 	if (err) {
 		nlmsg_free(msg);
@@ -1453,15 +1467,32 @@  static int devlink_nl_cmd_eswitch_mode_set_doit(struct sk_buff *skb,
 	struct devlink *devlink = info->user_ptr[0];
 	const struct devlink_ops *ops = devlink->ops;
 	u16 mode;
+	u8 inline_mode;
+	int err = 0;
 
-	if (!info->attrs[DEVLINK_ATTR_ESWITCH_MODE])
-		return -EINVAL;
+	if (!ops)
+		return -EOPNOTSUPP;
 
-	mode = nla_get_u16(info->attrs[DEVLINK_ATTR_ESWITCH_MODE]);
+	if (info->attrs[DEVLINK_ATTR_ESWITCH_MODE]) {
+		if (!ops->eswitch_mode_set)
+			return -EOPNOTSUPP;
+		mode = nla_get_u16(info->attrs[DEVLINK_ATTR_ESWITCH_MODE]);
+		err = ops->eswitch_mode_set(devlink, mode);
+		if (err)
+			return err;
+	}
 
-	if (ops && ops->eswitch_mode_set)
-		return ops->eswitch_mode_set(devlink, mode);
-	return -EOPNOTSUPP;
+	if (info->attrs[DEVLINK_ATTR_ESWITCH_INLINE_MODE]) {
+		if (!ops->eswitch_inline_mode_set)
+			return -EOPNOTSUPP;
+		inline_mode = nla_get_u8(
+				info->attrs[DEVLINK_ATTR_ESWITCH_INLINE_MODE]);
+		err = ops->eswitch_inline_mode_set(devlink, inline_mode);
+		if (err)
+			return err;
+	}
+
+	return 0;
 }
 
 static const struct nla_policy devlink_nl_policy[DEVLINK_ATTR_MAX + 1] = {
@@ -1478,6 +1509,7 @@  static const struct nla_policy devlink_nl_policy[DEVLINK_ATTR_MAX + 1] = {
 	[DEVLINK_ATTR_SB_THRESHOLD] = { .type = NLA_U32 },
 	[DEVLINK_ATTR_SB_TC_INDEX] = { .type = NLA_U16 },
 	[DEVLINK_ATTR_ESWITCH_MODE] = { .type = NLA_U16 },
+	[DEVLINK_ATTR_ESWITCH_INLINE_MODE] = { .type = NLA_U8 },
 };
 
 static const struct genl_ops devlink_nl_ops[] = {