Skip to contentSkip to conversation

XDP generic related fixes

PATCH Superseded Stephen Hemminger 3 participants 12 messages May 19, 2019 – May 21, 2019

Subsystems (4)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover XDP generic related fixes

This set of patches came about while investigating XDP generic on Azure. The split brain nature of the accelerated networking exposed issues with the stack device model.

The real fix is in the second patch which is a redo of earlier patch from Jason Wang.

v2 - hold off the comment fixes for net-next

Stephen Hemminger (2):
netvsc: unshare skb in VF rx handler
net: core: generic XDP support for stacked device

 drivers/net/hyperv/netvsc_drv.c |  6 ++++++
 net/core/dev.c                  | 10 ++++++++++
 2 files changed, 16 insertions(+)
Signature · 1 line
2.20.1
#1 lore raw

PATCH v2 1/2 netvsc: unshare skb in VF rx handler

The netvsc VF skb handler should make sure that skb is not shared. Similar logic already exists in bonding and team device drivers.

This is not an issue in practice because the VF devicex does not send up shared skb's. But the netvsc driver should do the right thing if it did.

Fixes 0c195567a8f6 ("netvsc: transparent VF management")
---
 drivers/net/hyperv/netvsc_drv.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/net/hyperv/netvsc_drv.c +6 -0
index 06393b215102..9873b8679f81 100644
@@ -2000,6 +2000,12 @@ static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
 	struct netvsc_vf_pcpu_stats *pcpu_stats
 		 = this_cpu_ptr(ndev_ctx->vf_stats);
 
+	skb = skb_share_check(skb, GFP_ATOMIC);
+	if (unlikely(!skb))
+		return RX_HANDLER_CONSUMED;
+
+	*pskb = skb;
+
 	skb->dev = ndev;
 
 	u64_stats_update_begin(&pcpu_stats->syncp);
Signature · 1 line
2.20.1
#2 lore raw

PATCH v2 2/2 net: core: generic XDP support for stacked device

In reply to Stephen Hemminger

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")
---
 net/core/dev.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
net/core/dev.c +10 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}
 			goto another_round;
 		case RX_HANDLER_EXACT:
 			deliver_exact = true;
Signature · 1 line
2.20.1
#3 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 28 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---
net/core/dev.c | 10 ++++++++++

1 file changed, 10 insertions(+)
net/core/dev.c +0 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4933

Quoted 10 lines
net/core/dev.c +10 -0
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

Quoted 6 lines
net/core/dev.c +0 -0
 			goto another_round;
 		case RX_HANDLER_EXACT:
 			deliver_exact = true;
Signature · 1 line
2.20.1
#4 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 19 lines · On Mon, 20 May 2019 11:11:05 +0200 Jiri Pirko <[email protected]> wrote:
Quoted 9 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

Quoted 8 lines

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---

Quoted 6 lines

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

Yes, this is a sensitive path. Another alternative is to fix it inside each device (netvsc). That is what my earlier patch did and that is what is being done now (probably will make it into the RHEL on Azure drivers).

Quoted 6 lines

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

That could work, but has the question about doing XDP farther down call stack (lower performance).

There is also the case what if both paths support XDP in driver. This would be the ideal case, how would this work?

#5 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 38 lines · Stephen Hemminger
Quoted 18 lines · On Mon, 20 May 2019 11:11:05 +0200 Jiri Pirko <[email protected]> wrote:
Quoted 9 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

Quoted 8 lines

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---

Quoted 6 lines

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

Yes, this is a sensitive path. Another alternative is to fix it inside each device (netvsc). That is what my earlier patch did and that is what is being done now (probably will make it into the RHEL on Azure drivers).

Quoted 6 lines

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

Quoted 7 lines

That could work, but has the question about doing XDP farther down call stack (lower performance).

There is also the case what if both paths support XDP in driver. This would be the ideal case, how would this work?

I think we have a clear request of allowing native XDP to work on stacked device. We're missing some generic building blocks (like XDP rx handler) here.

Thanks

#6 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

In reply to Jiri Pirko
Quoted 44 lines · On Mon, 20 May 2019 11:11:05 +0200 Jiri Pirko <[email protected]> wrote:
Quoted 28 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---
net/core/dev.c | 10 ++++++++++

1 file changed, 10 insertions(+)
net/core/dev.c +0 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4933

Quoted 10 lines
net/core/dev.c +10 -0
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}  

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

Quoted 9 lines

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

I am trying that now. But one problem is that it would break the case where XDP was being run on one leg of a bridge. For example if eth1 is part of br0; then it would no longer be possible to run XDP on eth1.

Running XDP on eth1 might be used to do some kind of ILA or overlay network. That change would break it.

#7 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 57 lines · Mon, May 20, 2019 at 06:04:05PM CEST, [email protected] wrote:
Quoted 44 lines · On Mon, 20 May 2019 11:11:05 +0200 Jiri Pirko <[email protected]> wrote:
Quoted 28 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---
net/core/dev.c | 10 ++++++++++

1 file changed, 10 insertions(+)
net/core/dev.c +0 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4933

Quoted 10 lines
net/core/dev.c +10 -0
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}  

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

Quoted 9 lines

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

Quoted 3 lines

I am trying that now. But one problem is that it would break the case where XDP was being run on one leg of a bridge. For example if eth1 is part of br0; then it would no longer be possible to run XDP on eth1.

I don't see why not. The xdp is still run in __netif_receive_skb_core() before goto another_round.

I was thinking about patch similar to this:

net/core/dev.c +13 -45
index b6b8505cfb3e..4c3fdda85544 100644
@@ -4502,23 +4502,6 @@ static int netif_rx_internal(struct sk_buff *skb)
 
 	trace_netif_rx(skb);
 
-	if (static_branch_unlikely(&generic_xdp_needed_key)) {
-		int ret;
-
-		preempt_disable();
-		rcu_read_lock();
-		ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
-		rcu_read_unlock();
-		preempt_enable();
-
-		/* Consider XDP consuming the packet a success from
-		 * the netdev point of view we do not want to count
-		 * this as an error.
-		 */
-		if (ret != XDP_PASS)
-			return NET_RX_SUCCESS;
-	}
-
 #ifdef CONFIG_RPS
 	if (static_branch_unlikely(&rps_needed)) {
 		struct rps_dev_flow voidflow, *rflow = &voidflow;
@@ -4858,6 +4841,19 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 
 	__this_cpu_inc(softnet_data.processed);
 
+	if (static_branch_unlikely(&generic_xdp_needed_key)) {
+		int ret2;
+
+		preempt_disable();
+		rcu_read_lock();
+		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
+		rcu_read_unlock();
+		preempt_enable();
+
+		if (ret2 != XDP_PASS)
+			return NET_RX_DROP;
+	}
+
 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
 	    skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
 		skb = skb_vlan_untag(skb);
@@ -5178,19 +5174,6 @@ static int netif_receive_skb_internal(struct sk_buff *skb)
 	if (skb_defer_rx_timestamp(skb))
 		return NET_RX_SUCCESS;
 
-	if (static_branch_unlikely(&generic_xdp_needed_key)) {
-		int ret;
-
-		preempt_disable();
-		rcu_read_lock();
-		ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
-		rcu_read_unlock();
-		preempt_enable();
-
-		if (ret != XDP_PASS)
-			return NET_RX_DROP;
-	}
-
 	rcu_read_lock();
 #ifdef CONFIG_RPS
 	if (static_branch_unlikely(&rps_needed)) {
@@ -5224,21 +5207,6 @@ static void netif_receive_skb_list_internal(struct list_head *head)
 	}
 	list_splice_init(&sublist, head);
 
-	if (static_branch_unlikely(&generic_xdp_needed_key)) {
-		preempt_disable();
-		rcu_read_lock();
-		list_for_each_entry_safe(skb, next, head, list) {
-			xdp_prog = rcu_dereference(skb->dev->xdp_prog);
-			skb_list_del_init(skb);
-			if (do_xdp_generic(xdp_prog, skb) == XDP_PASS)
-				list_add_tail(&skb->list, &sublist);
-		}
-		rcu_read_unlock();
-		preempt_enable();
-		/* Put passed packets back on main list */
-		list_splice_init(&sublist, head);
-	}
-
 	rcu_read_lock();
 #ifdef CONFIG_RPS
 	if (static_branch_unlikely(&rps_needed)) {
#8 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4856

Quoted 13 lines · On Tue, 21 May 2019 08:15:36 +0200 Jiri Pirko <[email protected]> wrote:
net/core/dev.c +13 -0
+	if (static_branch_unlikely(&generic_xdp_needed_key)) {
+		int ret2;
+
+		preempt_disable();
+		rcu_read_lock();
+		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
+		rcu_read_unlock();
+		preempt_enable();
+
+		if (ret2 != XDP_PASS)
+			return NET_RX_DROP;
+	}
+

rcu_read_lock is already held by callers of __netif_receive_skb_core

#9 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 18 lines · Tue, May 21, 2019 at 04:45:53PM CEST, [email protected] wrote:

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4856

Quoted 13 lines · On Tue, 21 May 2019 08:15:36 +0200 Jiri Pirko <[email protected]> wrote:
net/core/dev.c +13 -0
+	if (static_branch_unlikely(&generic_xdp_needed_key)) {
+		int ret2;
+
+		preempt_disable();
+		rcu_read_lock();
+		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
+		rcu_read_unlock();
+		preempt_enable();
+
+		if (ret2 != XDP_PASS)
+			return NET_RX_DROP;
+	}
+

rcu_read_lock is already held by callers of __netif_receive_skb_core

Sure, the purpose of the draft was just to show the idea.

#10 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

In reply to Jiri Pirko
Quoted 34 lines · Jiri Pirko
Quoted 33 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---
net/core/dev.c | 10 ++++++++++

1 file changed, 10 insertions(+)
net/core/dev.c +5 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
Quoted 8 lines

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4933

Quoted 5 lines
net/core/dev.c +5 -0
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

I may miss something, did you see any issue for bonding with this patch?

Quoted 3 lines

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

Do you mean we need do something similar after vlan_do_receive() returns true?

Quoted 5 lines

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

Probably just after another_round label. And this means generic XDP is done after RPS which could be even better.

Thanks

Quoted 9 lines
Quoted 9 lines
net/core/dev.c +0 -0
 			goto another_round;
 		case RX_HANDLER_EXACT:
 			deliver_exact = true;
Signature · 1 line
2.20.1
#11 lore raw

Re: [PATCH v2 net 2/2] net: core: generic XDP support for stacked device

Quoted 46 lines · Tue, May 21, 2019 at 06:47:23AM CEST, [email protected] wrote:
Quoted 34 lines · Jiri Pirko
Quoted 33 lines · Sun, May 19, 2019 at 05:10:46AM CEST, [email protected] wrote:

When a device is stacked like (team, bonding, failsafe or netvsc) the XDP generic program for the parent device is not called. In these cases, the rx handler changes skb->dev to its own in the receive handler, and returns RX_HANDLER_ANOTHER. Fix this by calling do_xdp_generic if necessary before starting another round.

Review of all the places RX_HANDLER_ANOTHER is returned show that the current devices do correctly change skb->dev.

There was an older patch that got abandoned that did the same thing, this is just a rewrite.

Fixes d445516966dc ("net: xdp: support xdp generic on virtual devices")

---
net/core/dev.c | 10 ++++++++++

1 file changed, 10 insertions(+)
net/core/dev.c +5 -0
index b6b8505cfb3e..240d0b2de1a8 100644
@@ -4921,6 +4921,16 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 			ret = NET_RX_SUCCESS;
 			goto out;
 		case RX_HANDLER_ANOTHER:
+			if (static_branch_unlikely(&generic_xdp_needed_key)) {
+				struct bpf_prog *xdp_prog;
+
+				xdp_prog = rcu_dereference(skb->dev->xdp_prog);
+				ret = do_xdp_generic(xdp_prog, skb);
Quoted 8 lines

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 4933

Quoted 5 lines
net/core/dev.c +5 -0
+				if (ret != XDP_PASS) {
+					ret = NET_RX_SUCCESS;
+					goto out;
+				}
+			}

I'm always scarred of changes like this. The history tells us that this codepaths are very fragile. It took us non-trivial efford to fix bonding here, not to mention vlans (that was pain).

Quoted 1 line

I may miss something, did you see any issue for bonding with this patch?

No, I was talking about past.

Quoted 9 lines
Quoted 5 lines

The reason for troubles was often fact that different flows were treated differently (vlan accel/non-accel).

Do you mean we need do something similar after vlan_do_receive() returns true?

No.

Quoted 11 lines
Quoted 7 lines

This patch calls do_xdp_generic for master device in different point in the receive patch comparing to lower device. Would it be possible to unify this? E.g. by moving do_xdp_generice() call from netif_rx_internal()/netif_receive_skb_internal() here,
to the beginning of __netif_receive_skb_core()?

Probably just after another_round label. And this means generic XDP is done after RPS which could be even better.

Yes. That is exactly the place I have in mind.

Quoted 13 lines

Thanks

Quoted 9 lines
Quoted 9 lines
net/core/dev.c +0 -0
 			goto another_round;
 		case RX_HANDLER_EXACT:
 			deliver_exact = true;
Signature · 1 line
2.20.1