[ovs-dev,v10,1/1] Avoid dp_hash recirculation for balance-tcp bond selection mode
diff mbox series

Message ID 1578460684-20228-2-git-send-email-vishal.deep.ajmera@ericsson.com
State New
Headers show
Series
  • Balance-tcp bond mode optimization
Related show

Commit Message

Jeff Squyres \(jsquyres\) via dev Jan. 8, 2020, 5:18 a.m. UTC
Problem:
--------
In OVS-DPDK, flows with output over a bond interface of type “balance-tcp”
(using a hash on TCP/UDP 5-tuple) get translated by the ofproto layer into
"HASH" and "RECIRC" datapath actions. After recirculation, the packet is
forwarded to the bond member port based on 8-bits of the datapath hash
value computed through dp_hash. This causes performance degradation in the
following ways:

1. The recirculation of the packet implies another lookup of the packet’s
flow key in the exact match cache (EMC) and potentially Megaflow classifier
(DPCLS). This is the biggest cost factor.

2. The recirculated packets have a new “RSS” hash and compete with the
original packets for the scarce number of EMC slots. This implies more
EMC misses and potentially EMC thrashing causing costly DPCLS lookups.

3. The 256 extra megaflow entries per bond for dp_hash bond selection put
additional load on the revalidation threads.

Owing to this performance degradation, deployments stick to “balance-slb”
bond mode even though it does not do active-active load balancing for
VXLAN- and GRE-tunnelled traffic because all tunnel packet have the same
source MAC address.

Proposed optimization:
----------------------
This proposal introduces a new load-balancing output action instead of
recirculation.

Maintain one table per-bond (could just be an array of uint16's) and
program it the same way internal flows are created today for each possible
hash value(256 entries) from ofproto layer. Use this table to load-balance
flows as part of output action processing.

Currently xlate_normal() -> output_normal() -> bond_update_post_recirc_rules()
-> bond_may_recirc() and compose_output_action__() generate
“dp_hash(hash_l4(0))” and “recirc(<RecircID>)” actions. In this case the
RecircID identifies the bond. For the recirculated packets the ofproto layer
installs megaflow entries that match on RecircID and masked dp_hash and send
them to the corresponding output port.

Instead, we will now generate action as
    "lb_output(bond,<bond id>)"

This combines hash computation (only if needed, else re-use RSS hash) and
inline load-balancing over the bond. This action is used *only* for balance-tcp
bonds in OVS-DPDK datapath (the OVS kernel datapath remains unchanged).

Example:
--------
Current scheme:
---------------
With 1 IP-UDP flow:

flow-dump from pmd on cpu core: 2
recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2828969, bytes:181054016, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)

recirc_id(0x1),dp_hash(0x113683bd/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:2828937, bytes:181051968, used:0.000s, actions:2

With 8 IP-UDP flows (with random UDP src port): (all hitting same DPCL):

flow-dump from pmd on cpu core: 2
recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)

recirc_id(0x1),dp_hash(0xf8e02b7e/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:377395, bytes:24153280, used:0.000s, actions:2
recirc_id(0x1),dp_hash(0xb236c260/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:333486, bytes:21343104, used:0.000s, actions:1
recirc_id(0x1),dp_hash(0x7d89eb18/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:348461, bytes:22301504, used:0.000s, actions:1
recirc_id(0x1),dp_hash(0xa78d75df/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:633353, bytes:40534592, used:0.000s, actions:2
recirc_id(0x1),dp_hash(0xb58d846f/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:319901, bytes:20473664, used:0.001s, actions:2
recirc_id(0x1),dp_hash(0x24534406/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:334985, bytes:21439040, used:0.001s, actions:1
recirc_id(0x1),dp_hash(0x3cf32550/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:326404, bytes:20889856, used:0.001s, actions:1

New scheme:
-----------
We can do with a single flow entry (for any number of new flows):

in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:lb_output(bond,1)

A new CLI has been added to dump datapath bond cache as given below.

“sudo ovs-appctl dpif-netdev/dp-bond-show [dp]”

root@ubuntu-190:performance_scripts # sudo ovs-appctl dpif-netdev/dp-bond-show
Bond cache:
        bond-id 1 :
                bucket 0 - slave 2
                bucket 1 - slave 1
                bucket 2 - slave 2
                bucket 3 - slave 1

Performance improvement:
------------------------
With a prototype of the proposed idea, the following perf improvement is seen
with Phy-VM-Phy UDP traffic, single flow. With multiple flows, the improvement
is even more enhanced (due to reduced number of flows).

1 VM:
*****
+--------------------------------------+
|                 mpps                 |
+--------------------------------------+
| Flows  master  with-opt.   %delta    |
+--------------------------------------+
| 1      4.47    5.73        28.21
| 10     4.17    5.35        28.45
| 400    3.52    5.39        53.01
| 1k     3.41    5.25        53.78
| 10k    2.53    4.57        80.44
| 100k   2.32    4.27        83.59
| 500k   2.32    4.27        83.59
+--------------------------------------+
mpps: million packets per second.
packet size 64 bytes.

Signed-off-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
Co-authored-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
Signed-off-by: Vishal Deep Ajmera <vishal.deep.ajmera@ericsson.com>

CC: Jan Scheurich <jan.scheurich@ericsson.com>
CC: Venkatesan Pradeep <venkatesan.pradeep@ericsson.com>
CC: Nitin Katiyar <nitin.katiyar@ericsson.com>
---
 datapath/linux/compat/include/linux/openvswitch.h |   1 +
 lib/dpif-netdev.c                                 | 502 ++++++++++++++++++++--
 lib/dpif-netlink.c                                |   3 +
 lib/dpif-provider.h                               |   9 +
 lib/dpif.c                                        |  49 +++
 lib/dpif.h                                        |   7 +
 lib/odp-execute.c                                 |   2 +
 lib/odp-util.c                                    |   4 +
 ofproto/bond.c                                    |  56 ++-
 ofproto/bond.h                                    |   9 +
 ofproto/ofproto-dpif-ipfix.c                      |   1 +
 ofproto/ofproto-dpif-sflow.c                      |   1 +
 ofproto/ofproto-dpif-xlate.c                      |  42 +-
 ofproto/ofproto-dpif.c                            |  24 ++
 ofproto/ofproto-dpif.h                            |   9 +-
 tests/lacp.at                                     |   9 +
 vswitchd/bridge.c                                 |   6 +
 vswitchd/vswitch.xml                              |  23 +
 18 files changed, 695 insertions(+), 62 deletions(-)

Comments

Ben Pfaff Jan. 8, 2020, 8:33 p.m. UTC | #1
On Wed, Jan 08, 2020 at 10:48:04AM +0530, Vishal Deep Ajmera wrote:
> Problem:
> --------
> In OVS-DPDK, flows with output over a bond interface of type “balance-tcp”
> (using a hash on TCP/UDP 5-tuple) get translated by the ofproto layer into
> "HASH" and "RECIRC" datapath actions. After recirculation, the packet is
> forwarded to the bond member port based on 8-bits of the datapath hash
> value computed through dp_hash. This causes performance degradation in the
> following ways:

Thanks for the patch!

I have a few minor stylistic suggestions, see below.

I'd like to hear Ilya's opinion on this.

diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 78f9cf928de2..c3a54e96346e 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -378,11 +378,12 @@ struct dp_netdev {
 
     struct conntrack *conntrack;
     struct pmd_auto_lb pmd_alb;
+
     /* Bonds.
      *
      * Any lookup into 'bonds' requires taking 'bond_mutex'. */
     struct ovs_mutex bond_mutex;
-    struct hmap bonds;
+    struct hmap bonds;          /* Contains "struct tx_bond"s. */
 };
 
 static void meter_lock(const struct dp_netdev *dp, uint32_t meter_id)
@@ -5581,10 +5582,7 @@ pmd_free_cached_bonds(struct dp_netdev_pmd_thread *pmd)
 
     /* Remove bonds from pmd which no longer exists. */
     HMAP_FOR_EACH_SAFE (bond, next, node, &pmd->bond_cache) {
-        struct tx_bond *tx = NULL;
-
-        tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);
-        if (!tx) {
+        if (!tx_bond_lookup(&pmd->tx_bonds, bond->bond_id)) {
             /* Bond no longer exist. Delete it from pmd. */
             hmap_remove(&pmd->bond_cache, &bond->node);
             free(bond);
@@ -5601,10 +5599,11 @@ pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
     pmd_free_cached_bonds(pmd);
     hmap_shrink(&pmd->bond_cache);
 
-    struct tx_bond *tx_bond, *tx_bond_cached;
+    struct tx_bond *tx_bond;
     HMAP_FOR_EACH (tx_bond, node, &pmd->tx_bonds) {
         /* Check if bond already exist on pmd. */
-        tx_bond_cached = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);
+        struct tx_bond *tx_bond_cached
+            = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);
 
         if (!tx_bond_cached) {
             /* Create new bond entry in cache. */
Ilya Maximets Jan. 9, 2020, 2:42 p.m. UTC | #2
On 08.01.2020 21:33, Ben Pfaff wrote:
> On Wed, Jan 08, 2020 at 10:48:04AM +0530, Vishal Deep Ajmera wrote:
>> Problem:
>> --------
>> In OVS-DPDK, flows with output over a bond interface of type “balance-tcp”
>> (using a hash on TCP/UDP 5-tuple) get translated by the ofproto layer into
>> "HASH" and "RECIRC" datapath actions. After recirculation, the packet is
>> forwarded to the bond member port based on 8-bits of the datapath hash
>> value computed through dp_hash. This causes performance degradation in the
>> following ways:
> 
> Thanks for the patch!
> 
> I have a few minor stylistic suggestions, see below.
> 
> I'd like to hear Ilya's opinion on this.

Thanks Ben for review!  I'll take another look at this patch in a next
couple of days.

> 
> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index 78f9cf928de2..c3a54e96346e 100644
> --- a/lib/dpif-netdev.c
> +++ b/lib/dpif-netdev.c
> @@ -378,11 +378,12 @@ struct dp_netdev {
>  
>      struct conntrack *conntrack;
>      struct pmd_auto_lb pmd_alb;
> +
>      /* Bonds.
>       *
>       * Any lookup into 'bonds' requires taking 'bond_mutex'. */
>      struct ovs_mutex bond_mutex;
> -    struct hmap bonds;
> +    struct hmap bonds;          /* Contains "struct tx_bond"s. */
>  };
>  
>  static void meter_lock(const struct dp_netdev *dp, uint32_t meter_id)
> @@ -5581,10 +5582,7 @@ pmd_free_cached_bonds(struct dp_netdev_pmd_thread *pmd)
>  
>      /* Remove bonds from pmd which no longer exists. */
>      HMAP_FOR_EACH_SAFE (bond, next, node, &pmd->bond_cache) {
> -        struct tx_bond *tx = NULL;
> -
> -        tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);
> -        if (!tx) {
> +        if (!tx_bond_lookup(&pmd->tx_bonds, bond->bond_id)) {
>              /* Bond no longer exist. Delete it from pmd. */
>              hmap_remove(&pmd->bond_cache, &bond->node);
>              free(bond);
> @@ -5601,10 +5599,11 @@ pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
>      pmd_free_cached_bonds(pmd);
>      hmap_shrink(&pmd->bond_cache);
>  
> -    struct tx_bond *tx_bond, *tx_bond_cached;
> +    struct tx_bond *tx_bond;
>      HMAP_FOR_EACH (tx_bond, node, &pmd->tx_bonds) {
>          /* Check if bond already exist on pmd. */
> -        tx_bond_cached = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);
> +        struct tx_bond *tx_bond_cached
> +            = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);
>  
>          if (!tx_bond_cached) {
>              /* Create new bond entry in cache. */
>
Jeff Squyres \(jsquyres\) via dev Jan. 20, 2020, 9 a.m. UTC | #3
> >
> > Thanks for the patch!
> >
> > I have a few minor stylistic suggestions, see below.
> >
> > I'd like to hear Ilya's opinion on this.
>
> Thanks Ben for review!  I'll take another look at this patch in a next
> couple of days.
>
> >

Hi Ilya, Ben,

Let me know if you have any comments. Can we get this patch reviewed/merged 
before the branching is done?

Warm Regards,
Vishal Ajmera
Ilya Maximets Jan. 20, 2020, 10:26 a.m. UTC | #4
On 20.01.2020 10:00, Vishal Deep Ajmera wrote:
>>>
>>> Thanks for the patch!
>>>
>>> I have a few minor stylistic suggestions, see below.
>>>
>>> I'd like to hear Ilya's opinion on this.
>>
>> Thanks Ben for review!  I'll take another look at this patch in a next
>> couple of days.
>>
>>>
> 
> Hi Ilya, Ben,
> 
> Let me know if you have any comments. Can we get this patch reviewed/merged 
> before the branching is done?

Sorry it takes so long.  Release time is very busy.
I'll try to look at this patch today.

Best regards, Ilya Maximets.
Ilya Maximets Jan. 20, 2020, 4:09 p.m. UTC | #5
On 08.01.2020 06:18, Vishal Deep Ajmera wrote:
> Problem:
> --------
> In OVS-DPDK, flows with output over a bond interface of type “balance-tcp”
> (using a hash on TCP/UDP 5-tuple) get translated by the ofproto layer into
> "HASH" and "RECIRC" datapath actions. After recirculation, the packet is
> forwarded to the bond member port based on 8-bits of the datapath hash
> value computed through dp_hash. This causes performance degradation in the
> following ways:
> 
> 1. The recirculation of the packet implies another lookup of the packet’s
> flow key in the exact match cache (EMC) and potentially Megaflow classifier
> (DPCLS). This is the biggest cost factor.
> 
> 2. The recirculated packets have a new “RSS” hash and compete with the
> original packets for the scarce number of EMC slots. This implies more
> EMC misses and potentially EMC thrashing causing costly DPCLS lookups.
> 
> 3. The 256 extra megaflow entries per bond for dp_hash bond selection put
> additional load on the revalidation threads.
> 
> Owing to this performance degradation, deployments stick to “balance-slb”
> bond mode even though it does not do active-active load balancing for
> VXLAN- and GRE-tunnelled traffic because all tunnel packet have the same
> source MAC address.
> 
> Proposed optimization:
> ----------------------
> This proposal introduces a new load-balancing output action instead of
> recirculation.
> 
> Maintain one table per-bond (could just be an array of uint16's) and
> program it the same way internal flows are created today for each possible
> hash value(256 entries) from ofproto layer. Use this table to load-balance
> flows as part of output action processing.
> 
> Currently xlate_normal() -> output_normal() -> bond_update_post_recirc_rules()
> -> bond_may_recirc() and compose_output_action__() generate
> “dp_hash(hash_l4(0))” and “recirc(<RecircID>)” actions. In this case the
> RecircID identifies the bond. For the recirculated packets the ofproto layer
> installs megaflow entries that match on RecircID and masked dp_hash and send
> them to the corresponding output port.
> 
> Instead, we will now generate action as
>     "lb_output(bond,<bond id>)"
> 
> This combines hash computation (only if needed, else re-use RSS hash) and
> inline load-balancing over the bond. This action is used *only* for balance-tcp
> bonds in OVS-DPDK datapath (the OVS kernel datapath remains unchanged).
> 
> Example:
> --------
> Current scheme:
> ---------------
> With 1 IP-UDP flow:
> 
> flow-dump from pmd on cpu core: 2
> recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2828969, bytes:181054016, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)
> 
> recirc_id(0x1),dp_hash(0x113683bd/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:2828937, bytes:181051968, used:0.000s, actions:2
> 
> With 8 IP-UDP flows (with random UDP src port): (all hitting same DPCL):
> 
> flow-dump from pmd on cpu core: 2
> recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)
> 
> recirc_id(0x1),dp_hash(0xf8e02b7e/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:377395, bytes:24153280, used:0.000s, actions:2
> recirc_id(0x1),dp_hash(0xb236c260/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:333486, bytes:21343104, used:0.000s, actions:1
> recirc_id(0x1),dp_hash(0x7d89eb18/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:348461, bytes:22301504, used:0.000s, actions:1
> recirc_id(0x1),dp_hash(0xa78d75df/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:633353, bytes:40534592, used:0.000s, actions:2
> recirc_id(0x1),dp_hash(0xb58d846f/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:319901, bytes:20473664, used:0.001s, actions:2
> recirc_id(0x1),dp_hash(0x24534406/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:334985, bytes:21439040, used:0.001s, actions:1
> recirc_id(0x1),dp_hash(0x3cf32550/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:326404, bytes:20889856, used:0.001s, actions:1
> 
> New scheme:
> -----------
> We can do with a single flow entry (for any number of new flows):
> 
> in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:lb_output(bond,1)
> 
> A new CLI has been added to dump datapath bond cache as given below.
> 
> “sudo ovs-appctl dpif-netdev/dp-bond-show [dp]”
> 
> root@ubuntu-190:performance_scripts # sudo ovs-appctl dpif-netdev/dp-bond-show
> Bond cache:
>         bond-id 1 :
>                 bucket 0 - slave 2
>                 bucket 1 - slave 1
>                 bucket 2 - slave 2
>                 bucket 3 - slave 1
> 
> Performance improvement:
> ------------------------
> With a prototype of the proposed idea, the following perf improvement is seen
> with Phy-VM-Phy UDP traffic, single flow. With multiple flows, the improvement
> is even more enhanced (due to reduced number of flows).
> 
> 1 VM:
> *****
> +--------------------------------------+
> |                 mpps                 |
> +--------------------------------------+
> | Flows  master  with-opt.   %delta    |
> +--------------------------------------+
> | 1      4.47    5.73        28.21
> | 10     4.17    5.35        28.45
> | 400    3.52    5.39        53.01
> | 1k     3.41    5.25        53.78
> | 10k    2.53    4.57        80.44
> | 100k   2.32    4.27        83.59
> | 500k   2.32    4.27        83.59
> +--------------------------------------+
> mpps: million packets per second.
> packet size 64 bytes.
> 
> Signed-off-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
> Co-authored-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
> Signed-off-by: Vishal Deep Ajmera <vishal.deep.ajmera@ericsson.com>
> 
> CC: Jan Scheurich <jan.scheurich@ericsson.com>
> CC: Venkatesan Pradeep <venkatesan.pradeep@ericsson.com>
> CC: Nitin Katiyar <nitin.katiyar@ericsson.com>
> ---
>  datapath/linux/compat/include/linux/openvswitch.h |   1 +
>  lib/dpif-netdev.c                                 | 502 ++++++++++++++++++++--
>  lib/dpif-netlink.c                                |   3 +
>  lib/dpif-provider.h                               |   9 +
>  lib/dpif.c                                        |  49 +++
>  lib/dpif.h                                        |   7 +
>  lib/odp-execute.c                                 |   2 +
>  lib/odp-util.c                                    |   4 +
>  ofproto/bond.c                                    |  56 ++-
>  ofproto/bond.h                                    |   9 +
>  ofproto/ofproto-dpif-ipfix.c                      |   1 +
>  ofproto/ofproto-dpif-sflow.c                      |   1 +
>  ofproto/ofproto-dpif-xlate.c                      |  42 +-
>  ofproto/ofproto-dpif.c                            |  24 ++
>  ofproto/ofproto-dpif.h                            |   9 +-
>  tests/lacp.at                                     |   9 +
>  vswitchd/bridge.c                                 |   6 +
>  vswitchd/vswitch.xml                              |  23 +
>  18 files changed, 695 insertions(+), 62 deletions(-)

I started reviewing, but stopped at some point because I realized
that current solution is over-complicated (too much of unnecessary
memory copies) and not thread safe.
See some comments inline.

I'll follow up with suggestions on how to implement this functionality
in a different way, since I don't think that it's possible to make
current version work correctly.

NACK for dpif-netdev part.

Best regards, Ilya Maximets.

> 
> diff --git a/datapath/linux/compat/include/linux/openvswitch.h b/datapath/linux/compat/include/linux/openvswitch.h
> index 2f0c655..7604a2a 100644
> --- a/datapath/linux/compat/include/linux/openvswitch.h
> +++ b/datapath/linux/compat/include/linux/openvswitch.h
> @@ -1021,6 +1021,7 @@ enum ovs_action_attr {
>  	OVS_ACTION_ATTR_TUNNEL_PUSH,   /* struct ovs_action_push_tnl*/
>  	OVS_ACTION_ATTR_TUNNEL_POP,    /* u32 port number. */
>  	OVS_ACTION_ATTR_DROP,          /* u32 xlate_error. */
> +	OVS_ACTION_ATTR_LB_OUTPUT,     /* u32 bond-id */

Please, add the period at the end of comment.

>  #endif
>  	__OVS_ACTION_ATTR_MAX,	      /* Nothing past this will be accepted
>  				       * from userspace. */
> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index 2421821..78f9cf9 100644
> --- a/lib/dpif-netdev.c
> +++ b/lib/dpif-netdev.c
> @@ -79,6 +79,7 @@
>  #include "unixctl.h"
>  #include "util.h"
>  #include "uuid.h"
> +#include "ofproto/bond.h"


This doesn't look good.  Please, avoid including 'ofproto' headers
from the 'lib' code.

>  
>  VLOG_DEFINE_THIS_MODULE(dpif_netdev);
>  
> @@ -377,6 +378,11 @@ struct dp_netdev {
>  
>      struct conntrack *conntrack;
>      struct pmd_auto_lb pmd_alb;

As Ben suggested, empty line here.

> +    /* Bonds.
> +     *
> +     * Any lookup into 'bonds' requires taking 'bond_mutex'. */
> +    struct ovs_mutex bond_mutex;
> +    struct hmap bonds;

As Ben suggested, comment here is required.

>  };
>  
>  static void meter_lock(const struct dp_netdev *dp, uint32_t meter_id)
> @@ -607,6 +613,20 @@ struct tx_port {
>      struct dp_netdev_rxq *output_pkts_rxqs[NETDEV_MAX_BURST];
>  };
>  
> +/* Contained by struct tx_bond 'slave_buckets' */
> +struct slave_entry {
> +    odp_port_t slave_id;
> +    atomic_ullong n_packets;
> +    atomic_ullong n_bytes;
> +};
> +
> +/* Contained by struct dp_netdev_pmd_thread's 'bond_cache' or 'tx_bonds'. */
> +struct tx_bond {
> +    struct hmap_node node;
> +    uint32_t bond_id;
> +    struct slave_entry slave_buckets[BOND_BUCKETS];
> +};
> +
>  /* A set of properties for the current processing loop that is not directly
>   * associated with the pmd thread itself, but with the packets being
>   * processed or the short-term system configuration (for example, time).
> @@ -718,7 +738,6 @@ struct dp_netdev_pmd_thread {
>      atomic_bool wait_for_reload;    /* Can we busy wait for the next reload? */
>      atomic_bool reload_tx_qid;      /* Do we need to reload static_tx_qid? */
>      atomic_bool exit;               /* For terminating the pmd thread. */
> -

Unrelated change.  Please, keep the line.

>      pthread_t thread;
>      unsigned core_id;               /* CPU core id of this pmd thread. */
>      int numa_id;                    /* numa node id of this pmd thread. */
> @@ -739,6 +758,11 @@ struct dp_netdev_pmd_thread {
>       * read by the pmd thread. */
>      struct hmap tx_ports OVS_GUARDED;
>  
> +    struct ovs_mutex bond_mutex;    /* Mutex for 'tx_bonds'. */
> +    /* Map of 'tx_bond's used for transmission.  Written by the main thread,
> +     * read/written by the pmd thread. */

Why PMD needs to write into this hash map?  Can we avoid this?
Sounds dangerous.

> +    struct hmap tx_bonds OVS_GUARDED;
> +
>      /* These are thread-local copies of 'tx_ports'.  One contains only tunnel
>       * ports (that support push_tunnel/pop_tunnel), the other contains ports
>       * with at least one txq (that support send).  A port can be in both.
> @@ -751,6 +775,8 @@ struct dp_netdev_pmd_thread {
>       * other instance will only be accessed by its own pmd thread. */
>      struct hmap tnl_port_cache;
>      struct hmap send_port_cache;
> +    /* These are thread-local copies of 'tx_bonds' */
> +    struct hmap bond_cache;
>  
>      /* Keep track of detailed PMD performance statistics. */
>      struct pmd_perf_stats perf_stats;
> @@ -830,6 +856,12 @@ static void dp_netdev_del_rxq_from_pmd(struct dp_netdev_pmd_thread *pmd,
>  static int
>  dp_netdev_pmd_flush_output_packets(struct dp_netdev_pmd_thread *pmd,
>                                     bool force);
> +static void dp_netdev_add_bond_tx_to_pmd(struct dp_netdev_pmd_thread *pmd,
> +                                         struct tx_bond *bond)
> +    OVS_REQUIRES(pmd->bond_mutex);
> +static void dp_netdev_del_bond_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
> +                                           struct tx_bond *tx)
> +    OVS_REQUIRES(pmd->bond_mutex);
>  
>  static void reconfigure_datapath(struct dp_netdev *dp)
>      OVS_REQUIRES(dp->port_mutex);
> @@ -838,6 +870,10 @@ static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd);
>  static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd);
>  static void pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd)
>      OVS_REQUIRES(pmd->port_mutex);
> +static void pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
> +    OVS_REQUIRES(pmd->bond_mutex);
> +static void pmd_load_bond_cache(struct dp_netdev_pmd_thread *pmd);
> +
>  static inline void
>  dp_netdev_pmd_try_optimize(struct dp_netdev_pmd_thread *pmd,
>                             struct polled_queue *poll_list, int poll_cnt);
> @@ -1396,6 +1432,45 @@ pmd_perf_show_cmd(struct unixctl_conn *conn, int argc,
>      par.command_type = PMD_INFO_PERF_SHOW;
>      dpif_netdev_pmd_info(conn, argc, argv, &par);
>  }
> +
> +static void
> +dpif_netdev_dp_bond_show(struct unixctl_conn *conn, int argc,
> +                         const char *argv[], void *aux OVS_UNUSED)
> +{
> +    struct ds reply = DS_EMPTY_INITIALIZER;
> +
> +    ovs_mutex_lock(&dp_netdev_mutex);
> +
> +    struct dp_netdev *dp = NULL;
> +    if (argc == 2) {
> +        dp = shash_find_data(&dp_netdevs, argv[1]);
> +    } else if (shash_count(&dp_netdevs) == 1) {
> +        /* There's only one datapath */
> +        dp = shash_first(&dp_netdevs)->data;
> +    }
> +    if (!dp) {
> +        ovs_mutex_unlock(&dp_netdev_mutex);
> +        unixctl_command_reply_error(conn,
> +                                    "please specify an existing datapath");
> +        return;
> +    }
> +    ds_put_cstr(&reply, "\nBonds:\n");
> +
> +    struct tx_bond *dp_bond_entry;
> +    HMAP_FOR_EACH (dp_bond_entry, node, &dp->bonds) {

This requires holding dp->bond_mutex.

> +        ds_put_format(&reply, "\tbond-id %u :\n",
> +                      dp_bond_entry->bond_id);
> +        for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
> +            ds_put_format(&reply, "\t\tbucket %u - slave %u \n",
> +                      bucket,
> +                      dp_bond_entry->slave_buckets[bucket].slave_id);
> +        }
> +    }
> +    ovs_mutex_unlock(&dp_netdev_mutex);
> +    unixctl_command_reply(conn, ds_cstr(&reply));
> +    ds_destroy(&reply);
> +}
> +
>  
>  static int
>  dpif_netdev_init(void)
> @@ -1427,6 +1502,9 @@ dpif_netdev_init(void)
>                               "[-us usec] [-q qlen]",
>                               0, 10, pmd_perf_log_set_cmd,
>                               NULL);
> +    unixctl_command_register("dpif-netdev/dp-bond-show", "[dp]",
> +                             0, 1, dpif_netdev_dp_bond_show,
> +                             NULL);
>      return 0;
>  }
>  
> @@ -1551,6 +1629,9 @@ create_dp_netdev(const char *name, const struct dpif_class *class,
>      ovs_mutex_init(&dp->port_mutex);
>      hmap_init(&dp->ports);
>      dp->port_seq = seq_create();
> +    ovs_mutex_init(&dp->bond_mutex);
> +    hmap_init(&dp->bonds);
> +
>      fat_rwlock_init(&dp->upcall_rwlock);
>  
>      dp->reconfigure_seq = seq_create();
> @@ -1674,6 +1755,14 @@ dp_netdev_free(struct dp_netdev *dp)
>      }
>      ovs_mutex_unlock(&dp->port_mutex);
>  
> +    ovs_mutex_lock(&dp->bond_mutex);
> +    struct tx_bond *bond, *next_bond;
> +    HMAP_FOR_EACH_SAFE (bond, next_bond, node, &dp->bonds) {
> +        hmap_remove(&dp->bonds, &bond->node);
> +        free(bond);
> +    }
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +
>      dp_netdev_destroy_all_pmds(dp, true);
>      cmap_destroy(&dp->poll_threads);
>  
> @@ -1692,6 +1781,9 @@ dp_netdev_free(struct dp_netdev *dp)
>      hmap_destroy(&dp->ports);
>      ovs_mutex_destroy(&dp->port_mutex);
>  
> +    hmap_destroy(&dp->bonds);
> +    ovs_mutex_destroy(&dp->bond_mutex);
> +
>      /* Upcalls must be disabled at this point */
>      dp_netdev_destroy_upcall_lock(dp);
>  
> @@ -1795,6 +1887,7 @@ dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd)
>          ovs_mutex_lock(&pmd->port_mutex);
>          pmd_load_cached_ports(pmd);
>          ovs_mutex_unlock(&pmd->port_mutex);
> +        pmd_load_bond_cache(pmd);
>          ovs_mutex_unlock(&pmd->dp->non_pmd_mutex);
>          return;
>      }
> @@ -1809,6 +1902,12 @@ hash_port_no(odp_port_t port_no)
>      return hash_int(odp_to_u32(port_no), 0);
>  }
>  
> +static uint32_t
> +hash_bond_id(uint32_t bond_id)
> +{
> +    return hash_int(bond_id, 0);
> +}
> +
>  static int
>  port_create(const char *devname, const char *type,
>              odp_port_t port_no, struct dp_netdev_port **portp)
> @@ -4349,6 +4448,19 @@ tx_port_lookup(const struct hmap *hmap, odp_port_t port_no)
>      return NULL;
>  }
>  
> +static struct tx_bond *
> +tx_bond_lookup(const struct hmap *hmap, uint32_t bond_id)
> +{
> +    struct tx_bond *tx;
> +
> +    HMAP_FOR_EACH_IN_BUCKET (tx, node, hash_bond_id(bond_id), hmap) {

Why not HMAP_FOR_EACH_WITH_HASH ?
BTW, it's better to calculate hash before the loop to be sure that
we're not recalculating it on each iteration.

> +        if (tx->bond_id == bond_id) {
> +            return tx;
> +        }
> +    }
> +    return NULL;
> +}
> +
>  static int
>  port_reconfigure(struct dp_netdev_port *port)
>  {
> @@ -4837,6 +4949,26 @@ pmd_remove_stale_ports(struct dp_netdev *dp,
>      ovs_mutex_unlock(&pmd->port_mutex);
>  }
>  
> +static void
> +pmd_remove_stale_bonds(struct dp_netdev *dp,
> +                       struct dp_netdev_pmd_thread *pmd)
> +    OVS_EXCLUDED(pmd->bond_mutex)
> +    OVS_EXCLUDED(dp->bond_mutex)
> +{
> +    ovs_mutex_lock(&dp->bond_mutex);
> +    ovs_mutex_lock(&pmd->bond_mutex);
> +
> +    struct tx_bond *tx, *tx_next;
> +    HMAP_FOR_EACH_SAFE (tx, tx_next, node, &pmd->tx_bonds) {
> +        if (!tx_bond_lookup(&dp->bonds, tx->bond_id)) {
> +            dp_netdev_del_bond_tx_from_pmd(pmd, tx);
> +        }
> +    }
> +
> +    ovs_mutex_unlock(&pmd->bond_mutex);
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +}
> +
>  /* Must be called each time a port is added/removed or the cmask changes.
>   * This creates and destroys pmd threads, reconfigures ports, opens their
>   * rxqs and assigns all rxqs/txqs to pmd threads. */
> @@ -4875,10 +5007,11 @@ reconfigure_datapath(struct dp_netdev *dp)
>          }
>      }
>  
> -    /* Remove from the pmd threads all the ports that have been deleted or
> -     * need reconfiguration. */
> +    /* Remove from the pmd threads all the ports/bonds that have been deleted
> +     * or need reconfiguration. */
>      CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
>          pmd_remove_stale_ports(dp, pmd);
> +        pmd_remove_stale_bonds(dp, pmd);
>      }
>  
>      /* Reload affected pmd threads.  We must wait for the pmd threads before
> @@ -5000,6 +5133,21 @@ reconfigure_datapath(struct dp_netdev *dp)
>          ovs_mutex_unlock(&pmd->port_mutex);
>      }
>  
> +    /* Add every bond to the tx cache of every pmd thread, if it's not
> +     * there already and if this pmd has at least one rxq to poll. */
> +    ovs_mutex_lock(&dp->bond_mutex);
> +    CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
> +        ovs_mutex_lock(&pmd->bond_mutex);
> +        if (hmap_count(&pmd->poll_list) || pmd->core_id == NON_PMD_CORE_ID) {
> +            struct tx_bond *bond;
> +            HMAP_FOR_EACH (bond, node, &dp->bonds) {
> +                dp_netdev_add_bond_tx_to_pmd(pmd, bond);
> +            }
> +        }
> +        ovs_mutex_unlock(&pmd->bond_mutex);
> +    }
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +
>      /* Reload affected pmd threads. */
>      reload_affected_pmds(dp);
>  
> @@ -5258,7 +5406,6 @@ pmd_rebalance_dry_run(struct dp_netdev *dp)
>      return ret;
>  }
>  
> -

Unrelated change.

>  /* Return true if needs to revalidate datapath flows. */
>  static bool
>  dpif_netdev_run(struct dpif *dpif)
> @@ -5428,6 +5575,56 @@ pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd)
>  }
>  
>  static void
> +pmd_free_cached_bonds(struct dp_netdev_pmd_thread *pmd)
> +{
> +    struct tx_bond *bond, *next;
> +
> +    /* Remove bonds from pmd which no longer exists. */
> +    HMAP_FOR_EACH_SAFE (bond, next, node, &pmd->bond_cache) {
> +        struct tx_bond *tx = NULL;
> +
> +        tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);

This lookup requires holding pmd->bond_mutex.

> +        if (!tx) {
> +            /* Bond no longer exist. Delete it from pmd. */
> +            hmap_remove(&pmd->bond_cache, &bond->node);
> +            free(bond);
> +        }
> +    }
> +}
> +
> +/* Copies bonds from 'pmd->tx_bonds' (shared with the main thread) to
> + * 'pmd->bond_cache' (thread local) */
> +static void
> +pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
> +    OVS_REQUIRES(pmd->bond_mutex)
> +{
> +    pmd_free_cached_bonds(pmd);
> +    hmap_shrink(&pmd->bond_cache);
> +
> +    struct tx_bond *tx_bond, *tx_bond_cached;
> +    HMAP_FOR_EACH (tx_bond, node, &pmd->tx_bonds) {
> +        /* Check if bond already exist on pmd. */
> +        tx_bond_cached = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);

You can't do a lookup in PMDs local cache.  Only PMD thread is allowed
to use it.

> +
> +        if (!tx_bond_cached) {
> +            /* Create new bond entry in cache. */
> +            tx_bond_cached = xmemdup(tx_bond, sizeof *tx_bond_cached);
> +            hmap_insert(&pmd->bond_cache, &tx_bond_cached->node,
> +                        hash_bond_id(tx_bond_cached->bond_id));
> +        } else {
> +            /* Update the slave-map. */
> +            for (int bucket = 0; bucket <= BOND_MASK; bucket++) {
> +                tx_bond_cached->slave_buckets[bucket].slave_id =
> +                    tx_bond->slave_buckets[bucket].slave_id;
> +            }
> +        }
> +        VLOG_DBG("Caching bond-id %d pmd %d",
> +                 tx_bond_cached->bond_id, pmd->core_id);
> +    }
> +}
> +
> +
> +static void
>  pmd_alloc_static_tx_qid(struct dp_netdev_pmd_thread *pmd)
>  {
>      ovs_mutex_lock(&pmd->dp->tx_qid_pool_mutex);
> @@ -5449,6 +5646,14 @@ pmd_free_static_tx_qid(struct dp_netdev_pmd_thread *pmd)
>      ovs_mutex_unlock(&pmd->dp->tx_qid_pool_mutex);
>  }
>  
> +static void
> +pmd_load_bond_cache(struct dp_netdev_pmd_thread *pmd)
> +{
> +    ovs_mutex_lock(&pmd->bond_mutex);
> +    pmd_load_cached_bonds(pmd);
> +    ovs_mutex_unlock(&pmd->bond_mutex);
> +}

Can we avoid introducing of this function by moving mutex acquaring inside
the 'pmd_load_cached_bonds'.  Function names are too similar. It's confusing.

> +
>  static int
>  pmd_load_queues_and_ports(struct dp_netdev_pmd_thread *pmd,
>                            struct polled_queue **ppoll_list)
> @@ -5476,6 +5681,8 @@ pmd_load_queues_and_ports(struct dp_netdev_pmd_thread *pmd,
>  
>      ovs_mutex_unlock(&pmd->port_mutex);
>  
> +    pmd_load_bond_cache(pmd);
> +
>      *ppoll_list = poll_list;
>      return i;
>  }
> @@ -6038,6 +6245,7 @@ dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp,
>      atomic_init(&pmd->reload, false);
>      ovs_mutex_init(&pmd->flow_mutex);
>      ovs_mutex_init(&pmd->port_mutex);
> +    ovs_mutex_init(&pmd->bond_mutex);
>      cmap_init(&pmd->flow_table);
>      cmap_init(&pmd->classifiers);
>      pmd->ctx.last_rxq = NULL;
> @@ -6048,6 +6256,8 @@ dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp,
>      hmap_init(&pmd->tx_ports);
>      hmap_init(&pmd->tnl_port_cache);
>      hmap_init(&pmd->send_port_cache);
> +    hmap_init(&pmd->tx_bonds);
> +    hmap_init(&pmd->bond_cache);
>      /* init the 'flow_cache' since there is no
>       * actual thread created for NON_PMD_CORE_ID. */
>      if (core_id == NON_PMD_CORE_ID) {
> @@ -6068,6 +6278,8 @@ dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd)
>      hmap_destroy(&pmd->send_port_cache);
>      hmap_destroy(&pmd->tnl_port_cache);
>      hmap_destroy(&pmd->tx_ports);
> +    hmap_destroy(&pmd->bond_cache);
> +    hmap_destroy(&pmd->tx_bonds);
>      hmap_destroy(&pmd->poll_list);
>      /* All flows (including their dpcls_rules) have been deleted already */
>      CMAP_FOR_EACH (cls, node, &pmd->classifiers) {
> @@ -6079,6 +6291,7 @@ dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd)
>      ovs_mutex_destroy(&pmd->flow_mutex);
>      seq_destroy(pmd->reload_seq);
>      ovs_mutex_destroy(&pmd->port_mutex);
> +    ovs_mutex_destroy(&pmd->bond_mutex);
>      free(pmd);
>  }
>  
> @@ -6232,6 +6445,45 @@ dp_netdev_del_port_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
>      free(tx);
>      pmd->need_reload = true;
>  }
> +
> +static void
> +dp_netdev_add_bond_tx_to_pmd(struct dp_netdev_pmd_thread *pmd,
> +                             struct tx_bond *bond)
> +    OVS_REQUIRES(pmd->bond_mutex)
> +{
> +    bool reload = false;
> +    struct tx_bond *tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);
> +    if (tx) {
> +        /* Check if mapping is changed. */
> +        for (int i = 0; i <= BOND_MASK; i++) {
> +            if (bond->slave_buckets[i].slave_id !=
> +                     tx->slave_buckets[i].slave_id) {
> +                /* Mapping is modified. Reload pmd bond cache again. */
> +                reload = true;
> +            }
> +            /* Copy the map always. */
> +            tx->slave_buckets[i].slave_id = bond->slave_buckets[i].slave_id;
> +        }
> +    } else {
> +        tx = xmemdup(bond, sizeof *tx);
> +        hmap_insert(&pmd->tx_bonds, &tx->node, hash_bond_id(bond->bond_id));
> +        reload = true;
> +    }
> +    if (reload) {
> +        pmd->need_reload = true;
> +    }
> +}
> +
> +/* Del 'tx' from the tx bond cache of 'pmd' */
> +static void
> +dp_netdev_del_bond_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
> +                               struct tx_bond *tx)
> +    OVS_REQUIRES(pmd->bond_mutex)
> +{
> +    hmap_remove(&pmd->tx_bonds, &tx->node);
> +    free(tx);
> +    pmd->need_reload = true;
> +}
>  
>  static char *
>  dpif_netdev_get_datapath_version(void)
> @@ -7007,6 +7259,13 @@ pmd_send_port_cache_lookup(const struct dp_netdev_pmd_thread *pmd,
>      return tx_port_lookup(&pmd->send_port_cache, port_no);
>  }
>  
> +static struct tx_bond *
> +pmd_tx_bond_cache_lookup(const struct dp_netdev_pmd_thread *pmd,
> +                         uint32_t bond_id)
> +{
> +    return tx_bond_lookup(&pmd->bond_cache, bond_id);
> +}
> +
>  static int
>  push_tnl_action(const struct dp_netdev_pmd_thread *pmd,
>                  const struct nlattr *attr,
> @@ -7057,6 +7316,91 @@ dp_execute_userspace_action(struct dp_netdev_pmd_thread *pmd,
>      }
>  }
>  
> +static bool
> +dp_execute_output_action(struct dp_netdev_pmd_thread *pmd,
> +                         struct dp_packet_batch *packets_,
> +                         bool should_steal, odp_port_t port_no)
> +{
> +    struct tx_port *p = pmd_send_port_cache_lookup(pmd, port_no);
> +    if (!OVS_LIKELY(p)) {

OVS_UNLIKELY(!p)

> +        return false;
> +    }
> +
> +    struct dp_packet_batch out;
> +    if (!should_steal) {
> +        dp_packet_batch_clone(&out, packets_);
> +        dp_packet_batch_reset_cutlen(packets_);
> +        packets_ = &out;
> +    }
> +    dp_packet_batch_apply_cutlen(packets_);
> +#ifdef DPDK_NETDEV
> +    if (OVS_UNLIKELY(!dp_packet_batch_is_empty(&p->output_pkts)
> +                     && packets_->packets[0]->source
> +                     != p->output_pkts.packets[0]->source)) {

Plaese, align this the same way as it was aligned before.

> +        /* netdev-dpdk assumes that all packets in a single
> +         * output batch has the same source. Flush here to
> +         * avoid memory access issues. */
> +        dp_netdev_pmd_flush_output_on_port(pmd, p);
> +    }
> +#endif
> +    if (dp_packet_batch_size(&p->output_pkts)
> +        + dp_packet_batch_size(packets_) > NETDEV_MAX_BURST) {
> +        /* Flush here to avoid overflow. */
> +        dp_netdev_pmd_flush_output_on_port(pmd, p);
> +    }
> +    if (dp_packet_batch_is_empty(&p->output_pkts)) {
> +        pmd->n_output_batches++;
> +    }
> +
> +    struct dp_packet *packet;
> +    DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
> +        p->output_pkts_rxqs[dp_packet_batch_size(&p->output_pkts)] =
> +            pmd->ctx.last_rxq;
> +        dp_packet_batch_add(&p->output_pkts, packet);
> +    }
> +    return true;
> +}
> +
> +static bool
> +dp_execute_lb_output_action(struct dp_netdev_pmd_thread *pmd,
> +                            struct dp_packet_batch *packets_,
> +                            bool should_steal, uint32_t bond)
> +{
> +    struct tx_bond *p_bond = pmd_tx_bond_cache_lookup(pmd, bond);
> +    if (!p_bond) {

OVS_UNLIKELY ?

> +        return false;
> +    }
> +
> +    struct dp_packet_batch del_pkts;
> +    dp_packet_batch_init(&del_pkts);
> +
> +    struct dp_packet *packet;
> +    DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
> +        /*
> +         * Lookup the bond-hash table using hash to get the slave.
> +         */
> +        uint32_t hash = dp_packet_get_rss_hash(packet);
> +        struct slave_entry *s_entry = &p_bond->slave_buckets[hash & BOND_MASK];
> +        odp_port_t bond_member = s_entry->slave_id;
> +        uint32_t size = dp_packet_size(packet);
> +
> +        struct dp_packet_batch output_pkt;
> +        dp_packet_batch_init_packet(&output_pkt, packet);
> +        if (OVS_LIKELY(dp_execute_output_action(pmd, &output_pkt, should_steal,
> +                                                bond_member))) {
> +            /* Update slave stats. */
> +            non_atomic_ullong_add(&s_entry->n_packets, 1);
> +            non_atomic_ullong_add(&s_entry->n_bytes, size);
> +        } else {
> +            dp_packet_batch_add(&del_pkts, packet);
> +        }
> +    }
> +
> +    /* Delete packets that failed OUTPUT action */
> +    dp_packet_delete_batch(&del_pkts, should_steal);

We have recently introduced drop stats.  No packets should be dropped
now without accounting.  Please, introduce new counter for that.
Details: a13a0209750c ("userspace: Improved packet drop statistics.")

> +    return true;
> +}
> +
>  static void
>  dp_execute_cb(void *aux_, struct dp_packet_batch *packets_,
>                const struct nlattr *a, bool should_steal)
> @@ -7072,43 +7416,15 @@ dp_execute_cb(void *aux_, struct dp_packet_batch *packets_,
>  
>      switch ((enum ovs_action_attr)type) {
>      case OVS_ACTION_ATTR_OUTPUT:
> -        p = pmd_send_port_cache_lookup(pmd, nl_attr_get_odp_port(a));
> -        if (OVS_LIKELY(p)) {
> -            struct dp_packet *packet;
> -            struct dp_packet_batch out;
> -
> -            if (!should_steal) {
> -                dp_packet_batch_clone(&out, packets_);
> -                dp_packet_batch_reset_cutlen(packets_);
> -                packets_ = &out;
> -            }
> -            dp_packet_batch_apply_cutlen(packets_);
> -
> -#ifdef DPDK_NETDEV
> -            if (OVS_UNLIKELY(!dp_packet_batch_is_empty(&p->output_pkts)
> -                             && packets_->packets[0]->source
> -                                != p->output_pkts.packets[0]->source)) {
> -                /* XXX: netdev-dpdk assumes that all packets in a single
> -                 *      output batch has the same source. Flush here to
> -                 *      avoid memory access issues. */
> -                dp_netdev_pmd_flush_output_on_port(pmd, p);
> -            }
> -#endif
> -            if (dp_packet_batch_size(&p->output_pkts)
> -                + dp_packet_batch_size(packets_) > NETDEV_MAX_BURST) {
> -                /* Flush here to avoid overflow. */
> -                dp_netdev_pmd_flush_output_on_port(pmd, p);
> -            }
> -
> -            if (dp_packet_batch_is_empty(&p->output_pkts)) {
> -                pmd->n_output_batches++;
> -            }
> +        if (dp_execute_output_action(pmd, packets_, should_steal,
> +                                     nl_attr_get_odp_port(a))) {
> +            return;
> +        }
> +        break;
>  
> -            DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
> -                p->output_pkts_rxqs[dp_packet_batch_size(&p->output_pkts)] =
> -                                                             pmd->ctx.last_rxq;
> -                dp_packet_batch_add(&p->output_pkts, packet);
> -            }
> +    case OVS_ACTION_ATTR_LB_OUTPUT:
> +        if (dp_execute_lb_output_action(pmd, packets_, should_steal,
> +                                        nl_attr_get_u32(a))) {
>              return;
>          } else {
>              COVERAGE_ADD(datapath_drop_invalid_port,

This doesn't look correct.  'else' branch is for OVS_ACTION_ATTR_OUTPUT.

> @@ -7666,6 +7982,109 @@ dpif_netdev_ipf_dump_done(struct dpif *dpif OVS_UNUSED, void *ipf_dump_ctx)
>  
>  }
>  
> +static int
> +dpif_netdev_bond_add(struct dpif *dpif, uint32_t bond_id,
> +                     odp_port_t slave_map[])
> +{
> +    struct dp_netdev *dp = get_dp_netdev(dpif);
> +
> +    ovs_mutex_lock(&dp->bond_mutex);
> +    /*
> +     * Lookup for the bond. If already exists, just update the slave-map.
> +     * Else create new.
> +     */
> +    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
> +    if (dp_bond_entry) {
> +        for (int bucket = 0; bucket <= BOND_MASK; bucket++) {
> +            dp_bond_entry->slave_buckets[bucket].slave_id = slave_map[bucket];
> +        }
> +
> +        struct dp_netdev_pmd_thread *pmd;
> +        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
> +            ovs_mutex_lock(&pmd->bond_mutex);
> +            dp_netdev_add_bond_tx_to_pmd(pmd, dp_bond_entry);
> +            ovs_mutex_unlock(&pmd->bond_mutex);
> +        }
> +    } else {
> +        struct tx_bond *dp_bond = xzalloc(sizeof *dp_bond);
> +        dp_bond->bond_id = bond_id;
> +        for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
> +            dp_bond->slave_buckets[bucket].slave_id = slave_map[bucket];
> +        }
> +        hmap_insert(&dp->bonds, &dp_bond->node,
> +                    hash_bond_id(dp_bond->bond_id));
> +
> +        /* Insert the bond map in all pmds. */
> +        struct dp_netdev_pmd_thread *pmd;
> +        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
> +            ovs_mutex_lock(&pmd->bond_mutex);
> +            dp_netdev_add_bond_tx_to_pmd(pmd, dp_bond);
> +            ovs_mutex_unlock(&pmd->bond_mutex);
> +        }
> +    }
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +    return 0;
> +}
> +
> +static int
> +dpif_netdev_bond_del(struct dpif *dpif, uint32_t bond_id)
> +{
> +    struct dp_netdev *dp = get_dp_netdev(dpif);
> +
> +    ovs_mutex_lock(&dp->bond_mutex);
> +
> +    /* Find the bond and delete it if present */
> +    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
> +    if (dp_bond_entry) {
> +        /* Remove the bond map in all pmds. */
> +        struct dp_netdev_pmd_thread *pmd;
> +        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
> +            ovs_mutex_lock(&pmd->bond_mutex);
> +            dp_netdev_del_bond_tx_from_pmd(pmd, dp_bond_entry);
> +            ovs_mutex_unlock(&pmd->bond_mutex);
> +        }
> +        hmap_remove(&dp->bonds, &dp_bond_entry->node);
> +        free(dp_bond_entry);
> +    }
> +
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +    return 0;
> +}
> +
> +static int
> +dpif_netdev_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
> +                           uint64_t *n_bytes)
> +{
> +    struct dp_netdev *dp = get_dp_netdev(dpif);
> +
> +    ovs_mutex_lock(&dp->bond_mutex);
> +
> +    /* Find the bond and retrieve stats if present */
> +    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
> +    if (dp_bond_entry) {
> +        /* Search the bond in all PMDs */
> +        struct dp_netdev_pmd_thread *pmd;
> +        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
> +            ovs_mutex_lock(&pmd->bond_mutex);
> +            struct tx_bond *pmd_bond_entry
> +                = tx_bond_lookup(&pmd->bond_cache, bond_id);

You can't access pmd->bond_cache.  Only the owning PMD thread allowed to do that.

> +            if (pmd_bond_entry) {
> +                /* Read bond stats. */
> +                for (int i = 0; i <= BOND_MASK; i++) {
> +                    uint64_t pmd_n_bytes;
> +                    atomic_read_relaxed(
> +                         &pmd_bond_entry->slave_buckets[i].n_bytes,
> +                         &pmd_n_bytes);
> +                    n_bytes[i] += pmd_n_bytes;
> +                }
> +            }
> +            ovs_mutex_unlock(&pmd->bond_mutex);
> +        }
> +    }
> +    ovs_mutex_unlock(&dp->bond_mutex);
> +    return 0;
> +}
> +
>  const struct dpif_class dpif_netdev_class = {
>      "netdev",
>      true,                       /* cleanup_required */
> @@ -7739,6 +8158,9 @@ const struct dpif_class dpif_netdev_class = {
>      dpif_netdev_meter_set,
>      dpif_netdev_meter_get,
>      dpif_netdev_meter_del,
> +    dpif_netdev_bond_add,
> +    dpif_netdev_bond_del,
> +    dpif_netdev_bond_stats_get,
>  };
>  
>  static void
> diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
> index 3d97c78..b1555df 100644
> --- a/lib/dpif-netlink.c
> +++ b/lib/dpif-netlink.c
> @@ -3995,6 +3995,9 @@ const struct dpif_class dpif_netlink_class = {
>      dpif_netlink_meter_set,
>      dpif_netlink_meter_get,
>      dpif_netlink_meter_del,
> +    NULL,                       /* bond_add */
> +    NULL,                       /* bond_del */
> +    NULL,                       /* bond_stats_get */
>  };
>  
>  static int
> diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h
> index b77317b..4415836 100644
> --- a/lib/dpif-provider.h
> +++ b/lib/dpif-provider.h
> @@ -616,6 +616,15 @@ struct dpif_class {
>       * zero. */
>      int (*meter_del)(struct dpif *, ofproto_meter_id meter_id,
>                       struct ofputil_meter_stats *, uint16_t n_bands);
> +
> +    /* Adds a bond with 'bond_id' and the slave-map to 'dpif'. */
> +    int (*bond_add)(struct dpif *dpif, uint32_t bond_id,
> +                    odp_port_t slave_map[]);
> +    /* Removes bond identified by 'bond_id' from 'dpif'. */
> +    int (*bond_del)(struct dpif *dpif, uint32_t bond_id);
> +    /* Reads bond stats from 'dpif'. */
> +    int (*bond_stats_get)(struct dpif *dpif, uint32_t bond_id,
> +                          uint64_t *n_bytes);
>  };
>  
>  extern const struct dpif_class dpif_netlink_class;
> diff --git a/lib/dpif.c b/lib/dpif.c
> index 955eaf6..41c0886 100644
> --- a/lib/dpif.c
> +++ b/lib/dpif.c
> @@ -1186,6 +1186,7 @@ dpif_execute_helper_cb(void *aux_, struct dp_packet_batch *packets_,
>  
>      case OVS_ACTION_ATTR_CT:
>      case OVS_ACTION_ATTR_OUTPUT:
> +    case OVS_ACTION_ATTR_LB_OUTPUT:
>      case OVS_ACTION_ATTR_TUNNEL_PUSH:
>      case OVS_ACTION_ATTR_TUNNEL_POP:
>      case OVS_ACTION_ATTR_USERSPACE:
> @@ -1236,6 +1237,7 @@ dpif_execute_helper_cb(void *aux_, struct dp_packet_batch *packets_,
>          struct dp_packet *clone = NULL;
>          uint32_t cutlen = dp_packet_get_cutlen(packet);
>          if (cutlen && (type == OVS_ACTION_ATTR_OUTPUT
> +                        || type == OVS_ACTION_ATTR_LB_OUTPUT
>                          || type == OVS_ACTION_ATTR_TUNNEL_PUSH
>                          || type == OVS_ACTION_ATTR_TUNNEL_POP
>                          || type == OVS_ACTION_ATTR_USERSPACE)) {
> @@ -1895,6 +1897,16 @@ dpif_supports_explicit_drop_action(const struct dpif *dpif)
>      return dpif_is_netdev(dpif);
>  }
>  
> +bool
> +dpif_supports_lb_output_action(const struct dpif *dpif)
> +{
> +    /*
> +     * Balance-tcp optimization is currently supported in netdev
> +     * datapath only.
> +     */
> +    return dpif_is_netdev(dpif);
> +}
> +
>  /* Meters */
>  void
>  dpif_meter_get_features(const struct dpif *dpif,
> @@ -1992,3 +2004,40 @@ dpif_meter_del(struct dpif *dpif, ofproto_meter_id meter_id,
>      }
>      return error;
>  }
> +
> +int
> +dpif_bond_add(struct dpif *dpif, uint32_t bond_id, odp_port_t slave_map[])
> +{
> +    int error = -1;
> +
> +    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_del) {
> +        error = dpif->dpif_class->bond_add(dpif, bond_id, slave_map);
> +    }
> +
> +    return error;
> +}
> +
> +int
> +dpif_bond_del(struct dpif *dpif, uint32_t bond_id)
> +{
> +    int error = -1;
> +
> +    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_del) {
> +        error = dpif->dpif_class->bond_del(dpif, bond_id);
> +    }
> +
> +    return error;
> +}
> +
> +int
> +dpif_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
> +                    uint64_t *n_bytes)
> +{
> +    int error = -1;
> +
> +    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_stats_get) {
> +        error = dpif->dpif_class->bond_stats_get(dpif, bond_id, n_bytes);
> +    }
> +
> +    return error;
> +}
> diff --git a/lib/dpif.h b/lib/dpif.h
> index a88851e..6b3216f 100644
> --- a/lib/dpif.h
> +++ b/lib/dpif.h
> @@ -908,6 +908,13 @@ char *dpif_get_dp_version(const struct dpif *);
>  bool dpif_supports_tnl_push_pop(const struct dpif *);
>  bool dpif_supports_explicit_drop_action(const struct dpif *);
>  
> +bool dpif_supports_lb_output_action(const struct dpif *);
> +
> +int dpif_bond_add(struct dpif *dpif, uint32_t bond_id, odp_port_t slave_map[]);
> +int dpif_bond_del(struct dpif *dpif, uint32_t bond_id);
> +int dpif_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
> +                        uint64_t *n_bytes);
> +
>  /* Log functions. */
>  struct vlog_module;
>  
> diff --git a/lib/odp-execute.c b/lib/odp-execute.c
> index 42d3335..6018e37 100644
> --- a/lib/odp-execute.c
> +++ b/lib/odp-execute.c
> @@ -793,6 +793,7 @@ requires_datapath_assistance(const struct nlattr *a)
>      switch (type) {
>          /* These only make sense in the context of a datapath. */
>      case OVS_ACTION_ATTR_OUTPUT:
> +    case OVS_ACTION_ATTR_LB_OUTPUT:
>      case OVS_ACTION_ATTR_TUNNEL_PUSH:
>      case OVS_ACTION_ATTR_TUNNEL_POP:
>      case OVS_ACTION_ATTR_USERSPACE:
> @@ -1068,6 +1069,7 @@ odp_execute_actions(void *dp, struct dp_packet_batch *batch, bool steal,
>              return;
>          }
>          case OVS_ACTION_ATTR_OUTPUT:
> +        case OVS_ACTION_ATTR_LB_OUTPUT:
>          case OVS_ACTION_ATTR_TUNNEL_PUSH:
>          case OVS_ACTION_ATTR_TUNNEL_POP:
>          case OVS_ACTION_ATTR_USERSPACE:
> diff --git a/lib/odp-util.c b/lib/odp-util.c
> index d6cb9e6..3472a8c 100644
> --- a/lib/odp-util.c
> +++ b/lib/odp-util.c
> @@ -119,6 +119,7 @@ odp_action_len(uint16_t type)
>  
>      switch ((enum ovs_action_attr) type) {
>      case OVS_ACTION_ATTR_OUTPUT: return sizeof(uint32_t);
> +    case OVS_ACTION_ATTR_LB_OUTPUT: return sizeof(uint32_t);
>      case OVS_ACTION_ATTR_TRUNC: return sizeof(struct ovs_action_trunc);
>      case OVS_ACTION_ATTR_TUNNEL_PUSH: return ATTR_LEN_VARIABLE;
>      case OVS_ACTION_ATTR_TUNNEL_POP: return sizeof(uint32_t);
> @@ -1122,6 +1123,9 @@ format_odp_action(struct ds *ds, const struct nlattr *a,
>      case OVS_ACTION_ATTR_OUTPUT:
>          odp_portno_name_format(portno_names, nl_attr_get_odp_port(a), ds);
>          break;
> +    case OVS_ACTION_ATTR_LB_OUTPUT:
> +        ds_put_format(ds, "lb_output(bond,%"PRIu32")", nl_attr_get_u32(a));
> +        break;
>      case OVS_ACTION_ATTR_TRUNC: {
>          const struct ovs_action_trunc *trunc =
>                         nl_attr_get_unspec(a, sizeof *trunc);
> diff --git a/ofproto/bond.c b/ofproto/bond.c
> index 405202f..d540635 100644
> --- a/ofproto/bond.c
> +++ b/ofproto/bond.c
> @@ -54,10 +54,6 @@ static struct ovs_rwlock rwlock = OVS_RWLOCK_INITIALIZER;
>  static struct hmap all_bonds__ = HMAP_INITIALIZER(&all_bonds__);
>  static struct hmap *const all_bonds OVS_GUARDED_BY(rwlock) = &all_bonds__;
>  
> -/* Bit-mask for hashing a flow down to a bucket. */
> -#define BOND_MASK 0xff
> -#define BOND_BUCKETS (BOND_MASK + 1)
> -
>  /* Priority for internal rules created to handle recirculation */
>  #define RECIRC_RULE_PRIORITY 20
>  
> @@ -126,6 +122,8 @@ struct bond {
>      enum lacp_status lacp_status; /* Status of LACP negotiations. */
>      bool bond_revalidate;       /* True if flows need revalidation. */
>      uint32_t basis;             /* Basis for flow hash function. */
> +    bool use_bond_cache;        /* Use bond cache to avoid recirculation.
> +                                   Applicable only for Balance TCP mode. */
>  
>      /* SLB specific bonding info. */
>      struct bond_entry *hash;     /* An array of BOND_BUCKETS elements. */
> @@ -185,7 +183,7 @@ static struct bond_slave *choose_output_slave(const struct bond *,
>                                                struct flow_wildcards *,
>                                                uint16_t vlan)
>      OVS_REQ_RDLOCK(rwlock);
> -static void update_recirc_rules__(struct bond *bond);
> +static void update_recirc_rules__(struct bond *bond, uint32_t bond_recirc_id);
>  static bool bond_is_falling_back_to_ab(const struct bond *);
>  
>  /* Attempts to parse 's' as the name of a bond balancing mode.  If successful,
> @@ -262,6 +260,7 @@ void
>  bond_unref(struct bond *bond)
>  {
>      struct bond_slave *slave;
> +    uint32_t bond_recirc_id = 0;
>  
>      if (!bond || ovs_refcount_unref_relaxed(&bond->ref_cnt) != 1) {
>          return;
> @@ -282,12 +281,13 @@ bond_unref(struct bond *bond)
>  
>      /* Free bond resources. Remove existing post recirc rules. */
>      if (bond->recirc_id) {
> +        bond_recirc_id = bond->recirc_id;
>          recirc_free_id(bond->recirc_id);
>          bond->recirc_id = 0;
>      }
>      free(bond->hash);
>      bond->hash = NULL;
> -    update_recirc_rules__(bond);
> +    update_recirc_rules__(bond, bond_recirc_id);
>  
>      hmap_destroy(&bond->pr_rule_ops);
>      free(bond->name);
> @@ -328,13 +328,14 @@ add_pr_rule(struct bond *bond, const struct match *match,
>   * lock annotation. Currently, only 'bond_unref()' calls
>   * this function directly.  */
>  static void
> -update_recirc_rules__(struct bond *bond)
> +update_recirc_rules__(struct bond *bond, uint32_t bond_recirc_id)
>  {
>      struct match match;
>      struct bond_pr_rule_op *pr_op, *next_op;
>      uint64_t ofpacts_stub[128 / 8];
>      struct ofpbuf ofpacts;
>      int i;
> +    ofp_port_t slave_map[BOND_MASK];
>  
>      ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
>  
> @@ -353,8 +354,18 @@ update_recirc_rules__(struct bond *bond)
>  
>                  add_pr_rule(bond, &match, slave->ofp_port,
>                              &bond->hash[i].pr_rule);
> +                slave_map[i] = slave->ofp_port;
> +            } else {
> +                slave_map[i] = OFP_PORT_C(-1);
>              }
>          }
> +        if (bond->ofproto->backer->rt_support.lb_output_action) {
> +            ofproto_dpif_bundle_add(bond->ofproto, bond->recirc_id, slave_map);
> +        }
> +    } else {
> +        if (bond->ofproto->backer->rt_support.lb_output_action) {
> +            ofproto_dpif_bundle_del(bond->ofproto, bond_recirc_id);
> +        }
>      }
>  
>      HMAP_FOR_EACH_SAFE(pr_op, next_op, hmap_node, &bond->pr_rule_ops) {
> @@ -404,7 +415,7 @@ static void
>  update_recirc_rules(struct bond *bond)
>      OVS_REQ_RDLOCK(rwlock)
>  {
> -    update_recirc_rules__(bond);
> +    update_recirc_rules__(bond, bond->recirc_id);
>  }
>  
>  /* Updates 'bond''s overall configuration to 's'.
> @@ -467,6 +478,10 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s)
>          recirc_free_id(bond->recirc_id);
>          bond->recirc_id = 0;
>      }
> +    if (bond->use_bond_cache != s->use_bond_cache) {
> +        bond->use_bond_cache = s->use_bond_cache;
> +        revalidate = true;
> +    }
>  
>      if (bond->balance == BM_AB || !bond->hash || revalidate) {
>          bond_entry_reset(bond);
> @@ -944,6 +959,15 @@ bond_recirculation_account(struct bond *bond)
>      OVS_REQ_WRLOCK(rwlock)
>  {
>      int i;
> +    uint64_t n_bytes[BOND_BUCKETS] = {0};
> +
> +    if (bond->hash && bond->recirc_id) {
> +        /* Retrieve bond stats from datapath. */
> +        if (bond->ofproto->backer->rt_support.lb_output_action) {
> +            dpif_bond_stats_get(bond->ofproto->backer->dpif,
> +                                bond->recirc_id, n_bytes);
> +       }
> +    }
>  
>      for (i=0; i<=BOND_MASK; i++) {
>          struct bond_entry *entry = &bond->hash[i];
> @@ -953,8 +977,12 @@ bond_recirculation_account(struct bond *bond)
>              struct pkt_stats stats;
>              long long int used OVS_UNUSED;
>  
> -            rule->ofproto->ofproto_class->rule_get_stats(
> -                rule, &stats, &used);
> +            if (!bond->ofproto->backer->rt_support.lb_output_action) {
> +                rule->ofproto->ofproto_class->rule_get_stats(
> +                    rule, &stats, &used);
> +            } else {
> +                stats.n_bytes = n_bytes[i];
> +            }
>              bond_entry_account(entry, stats.n_bytes);
>          }
>      }
> @@ -1365,6 +1393,8 @@ bond_print_details(struct ds *ds, const struct bond *bond)
>                    may_recirc ? "yes" : "no", may_recirc ? recirc_id: -1);
>  
>      ds_put_format(ds, "bond-hash-basis: %"PRIu32"\n", bond->basis);
> +    ds_put_format(ds, "lb-output-action: %s\n",
> +                  bond->use_bond_cache ? "enabled" : "disabled");
>  
>      ds_put_format(ds, "updelay: %d ms\n", bond->updelay);
>      ds_put_format(ds, "downdelay: %d ms\n", bond->downdelay);
> @@ -1942,3 +1972,9 @@ bond_get_changed_active_slave(const char *name, struct eth_addr *mac,
>  
>      return false;
>  }
> +
> +bool
> +bond_get_cache_mode(const struct bond *bond)
> +{
> +    return bond->use_bond_cache;
> +}
> diff --git a/ofproto/bond.h b/ofproto/bond.h
> index e7c3d9b..88a4de1 100644
> --- a/ofproto/bond.h
> +++ b/ofproto/bond.h
> @@ -22,6 +22,10 @@
>  #include "ofproto-provider.h"
>  #include "packets.h"
>  
> +/* Bit-mask for hashing a flow down to a bucket. */
> +#define BOND_MASK 0xff
> +#define BOND_BUCKETS (BOND_MASK + 1)
> +
>  struct flow;
>  struct netdev;
>  struct ofpbuf;
> @@ -58,6 +62,8 @@ struct bond_settings {
>                                  /* The MAC address of the interface
>                                     that was active during the last
>                                     ovs run. */
> +    bool use_bond_cache;        /* Use bond cache. Only applicable for
> +                                   bond mode BALANCE TCP. */
>  };
>  
>  /* Program startup. */
> @@ -122,4 +128,7 @@ void bond_rebalance(struct bond *);
>  */
>  void bond_update_post_recirc_rules(struct bond *, uint32_t *recirc_id,
>                                     uint32_t *hash_basis);
> +
> +bool bond_get_cache_mode(const struct bond *);
> +
>  #endif /* bond.h */
> diff --git a/ofproto/ofproto-dpif-ipfix.c b/ofproto/ofproto-dpif-ipfix.c
> index b413768..5030978 100644
> --- a/ofproto/ofproto-dpif-ipfix.c
> +++ b/ofproto/ofproto-dpif-ipfix.c
> @@ -3017,6 +3017,7 @@ dpif_ipfix_read_actions(const struct flow *flow,
>          case OVS_ACTION_ATTR_CHECK_PKT_LEN:
>          case OVS_ACTION_ATTR_UNSPEC:
>          case OVS_ACTION_ATTR_DROP:
> +        case OVS_ACTION_ATTR_LB_OUTPUT:
>          case __OVS_ACTION_ATTR_MAX:
>          default:
>              break;
> diff --git a/ofproto/ofproto-dpif-sflow.c b/ofproto/ofproto-dpif-sflow.c
> index f9ea47a..21dafa6 100644
> --- a/ofproto/ofproto-dpif-sflow.c
> +++ b/ofproto/ofproto-dpif-sflow.c
> @@ -1177,6 +1177,7 @@ dpif_sflow_read_actions(const struct flow *flow,
>          case OVS_ACTION_ATTR_CT:
>      case OVS_ACTION_ATTR_CT_CLEAR:
>          case OVS_ACTION_ATTR_METER:
> +        case OVS_ACTION_ATTR_LB_OUTPUT:
>              break;
>  
>          case OVS_ACTION_ATTR_SET_MASKED:
> diff --git a/ofproto/ofproto-dpif-xlate.c b/ofproto/ofproto-dpif-xlate.c
> index 4407f9c..0446387 100644
> --- a/ofproto/ofproto-dpif-xlate.c
> +++ b/ofproto/ofproto-dpif-xlate.c
> @@ -409,6 +409,8 @@ struct xlate_ctx {
>      struct ofpbuf action_set;   /* Action set. */
>  
>      enum xlate_error error;     /* Translation failed. */
> +
> +    bool tnl_push_no_recirc;    /* Tunnel push recirculation status */
>  };
>  
>  /* Structure to track VLAN manipulation */
> @@ -3727,12 +3729,16 @@ native_tunnel_output(struct xlate_ctx *ctx, const struct xport *xport,
>          ctx->xin->allow_side_effects = backup_side_effects;
>          ctx->xin->packet = backup_packet;
>          ctx->wc = backup_wc;
> +
> +        ctx->tnl_push_no_recirc = true;
>      } else {
>          /* In order to maintain accurate stats, use recirc for
>           * natvie tunneling.  */
>          nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, 0);
>          nl_msg_end_nested(ctx->odp_actions, clone_ofs);
> -    }
> +
> +        ctx->tnl_push_no_recirc = false;
> +   }
>  
>      /* Restore the flows after the translation. */
>      memcpy(&ctx->xin->flow, &old_flow, sizeof ctx->xin->flow);
> @@ -4183,7 +4189,6 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
>          xlate_commit_actions(ctx);
>  
>          if (xr) {
> -            /* Recirculate the packet. */
>              struct ovs_action_hash *act_hash;
>  
>              /* Hash action. */
> @@ -4192,15 +4197,30 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
>                  /* Algorithm supported by all datapaths. */
>                  hash_alg = OVS_HASH_ALG_L4;
>              }
> -            act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
> +
> +            if ((xr->hash_alg == OVS_HASH_ALG_L4) &&
> +                (ctx->xbridge->support.lb_output_action) &&
> +                !ctx->tnl_push_no_recirc &&
> +                bond_get_cache_mode(xport->xbundle->bond)) {
> +                /*
> +                 * If bond mode is balance-tcp and optimize balance tcp
> +                 * is enabled then use the hash directly for slave selection
> +                 * and avoid recirculation.
> +                 *
> +                 * Currently support for netdev datapath only.
> +                 */
> +                nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_LB_OUTPUT,
> +                               xr->recirc_id);
> +            } else {
> +                act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
>                                                  OVS_ACTION_ATTR_HASH,
>                                                  sizeof *act_hash);
> -            act_hash->hash_alg = hash_alg;
> -            act_hash->hash_basis = xr->hash_basis;
> -
> -            /* Recirc action. */
> -            nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
> -                           xr->recirc_id);
> +                act_hash->hash_alg = hash_alg;
> +                act_hash->hash_basis = xr->hash_basis;
> +                /* Recirculate the packet. */
> +                nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
> +                               xr->recirc_id);
> +            }
>          } else if (is_native_tunnel) {
>              /* Output to native tunnel port. */
>              native_tunnel_output(ctx, xport, flow, odp_port, truncate);
> @@ -7261,7 +7281,8 @@ count_output_actions(const struct ofpbuf *odp_actions)
>      int n = 0;
>  
>      NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
> -        if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
> +        if ((a->nla_type == OVS_ACTION_ATTR_OUTPUT) ||
> +            (a->nla_type == OVS_ACTION_ATTR_LB_OUTPUT)) {
>              n++;
>          }
>      }
> @@ -7458,6 +7479,7 @@ xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
>  
>          .action_set_has_group = false,
>          .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
> +        .tnl_push_no_recirc = false,
>      };
>  
>      /* 'base_flow' reflects the packet as it came in, but we need it to reflect
> diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
> index f782f86..3c5438b 100644
> --- a/ofproto/ofproto-dpif.c
> +++ b/ofproto/ofproto-dpif.c
> @@ -1580,6 +1580,8 @@ check_support(struct dpif_backer *backer)
>      backer->rt_support.ct_timeout = check_ct_timeout_policy(backer);
>      backer->rt_support.explicit_drop_action =
>          dpif_supports_explicit_drop_action(backer->dpif);
> +    backer->rt_support.lb_output_action=
> +        dpif_supports_lb_output_action(backer->dpif);
>  
>      /* Flow fields. */
>      backer->rt_support.odp.ct_state = check_ct_state(backer);
> @@ -3433,6 +3435,27 @@ bundle_remove(struct ofport *port_)
>      }
>  }
>  
> +int
> +ofproto_dpif_bundle_add(struct ofproto_dpif *ofproto,
> +                        uint32_t bond_id, const ofp_port_t slave_map[])
> +{
> +    /* Convert ofp_port to odp_port */
> +    odp_port_t odp_map[BOND_BUCKETS];
> +    for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
> +        odp_map[bucket] = (slave_map[bucket] == OFP_PORT_C(-1)
> +                           ? ODP_PORT_C(-1)
> +                           : ofp_port_to_odp_port(ofproto, slave_map[bucket]));
> +    }
> +
> +    return dpif_bond_add(ofproto->backer->dpif, bond_id, odp_map);
> +}
> +
> +int
> +ofproto_dpif_bundle_del(struct ofproto_dpif *ofproto, uint32_t bond_id)
> +{
> +    return dpif_bond_del(ofproto->backer->dpif, bond_id);
> +}
> +
>  static void
>  send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
>  {
> @@ -5563,6 +5586,7 @@ get_datapath_cap(const char *datapath_type, struct smap *cap)
>      smap_add(cap, "ct_timeout", s.ct_timeout ? "true" : "false");
>      smap_add(cap, "explicit_drop_action",
>               s.explicit_drop_action ? "true" :"false");
> +    smap_add(cap, "lb_output_action", s.lb_output_action ? "true" : "false");
>  }
>  
>  /* Gets timeout policy name in 'backer' based on 'zone', 'dl_type' and
> diff --git a/ofproto/ofproto-dpif.h b/ofproto/ofproto-dpif.h
> index c9d5df3..384a269 100644
> --- a/ofproto/ofproto-dpif.h
> +++ b/ofproto/ofproto-dpif.h
> @@ -202,8 +202,10 @@ struct group_dpif *group_dpif_lookup(struct ofproto_dpif *,
>      DPIF_SUPPORT_FIELD(bool, ct_timeout, "Conntrack timeout policy")        \
>                                                                              \
>      /* True if the datapath supports explicit drop action. */               \
> -    DPIF_SUPPORT_FIELD(bool, explicit_drop_action, "Explicit Drop action")
> -
> +    DPIF_SUPPORT_FIELD(bool, explicit_drop_action, "Explicit Drop action")  \
> +                                                                            \
> +    /* True if the datapath supports balance_tcp optimization */            \
> +    DPIF_SUPPORT_FIELD(bool, lb_output_action, "Optimized Balance TCP mode")
>  
>  /* Stores the various features which the corresponding backer supports. */
>  struct dpif_backer_support {
> @@ -379,6 +381,9 @@ int ofproto_dpif_add_internal_flow(struct ofproto_dpif *,
>                                     struct rule **rulep);
>  int ofproto_dpif_delete_internal_flow(struct ofproto_dpif *, struct match *,
>                                        int priority);
> +int ofproto_dpif_bundle_add(struct ofproto_dpif *, uint32_t bond_id,
> +                            const ofp_port_t slave_map[]);
> +int ofproto_dpif_bundle_del(struct ofproto_dpif *, uint32_t bond_id);
>  
>  bool ovs_native_tunneling_is_on(struct ofproto_dpif *);
>  
> diff --git a/tests/lacp.at b/tests/lacp.at
> index 7b460d7..c881a0b 100644
> --- a/tests/lacp.at
> +++ b/tests/lacp.at
> @@ -121,6 +121,7 @@ AT_CHECK([ovs-appctl bond/show], [0], [dnl
>  bond_mode: active-backup
>  bond may use recirculation: no, Recirc-ID : -1
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -286,6 +287,7 @@ slave: p3: current attached
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -301,6 +303,7 @@ slave p1: enabled
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -423,6 +426,7 @@ slave: p3: current attached
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -440,6 +444,7 @@ slave p1: enabled
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -555,6 +560,7 @@ slave: p3: current attached
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -572,6 +578,7 @@ slave p1: enabled
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -692,6 +699,7 @@ slave: p3: current attached
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> @@ -709,6 +717,7 @@ slave p1: enabled
>  bond_mode: balance-tcp
>  bond may use recirculation: yes, <del>
>  bond-hash-basis: 0
> +lb-output-action: disabled
>  updelay: 0 ms
>  downdelay: 0 ms
>  lacp_status: negotiated
> diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
> index 86c7b10..d4f34ad 100644
> --- a/vswitchd/bridge.c
> +++ b/vswitchd/bridge.c
> @@ -4589,6 +4589,12 @@ port_configure_bond(struct port *port, struct bond_settings *s)
>          /* OVSDB did not store the last active interface */
>          s->active_slave_mac = eth_addr_zero;
>      }
> +    /* Default set bond cache use to false. */
> +    s->use_bond_cache = false;
> +    if (s->balance == BM_TCP) {
> +        s->use_bond_cache = smap_get_bool(&port->cfg->other_config,
> +                                        "lb-output-action", false);
> +    }
>  }
>  
>  /* Returns true if 'port' is synthetic, that is, if we constructed it locally
> diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
> index 0ec726c..bdc7ef4 100644
> --- a/vswitchd/vswitch.xml
> +++ b/vswitchd/vswitch.xml
> @@ -1962,6 +1962,16 @@
>          <code>active-backup</code>.
>        </column>
>  
> +      <column name="other_config" key="lb-output-action"
> +              type='{"type": "boolean"}'>
> +        Enable/disable usage of optimized lb-output-action for
> +        balancing flows among output slaves in load balanced bonds in
> +        <code>balance-tcp</code>. When enabled, it uses optimized path for
> +        balance-tcp mode by using rss hash and avoids recirculation.
> +        It affects only new flows, i.e, existing flows remain unchanged.
> +        This knob does not affect other balancing modes.
> +      </column>
> +
>        <group title="Link Failure Detection">
>          <p>
>            An important part of link bonding is detecting that links are down so
> @@ -5722,6 +5732,19 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
>          higher performance for MPLS and active-active load balancing
>          bonding modes.
>        </column>
> +      <column name="capabilities" key="lb_output_action"
> +              type='{"type": "boolean"}'>
> +        If this is true, then the datapath supports optimized balance-tcp
> +        bond mode. This capability replaces existing <code>hash</code> and
> +        <code>recirc</code> actions with new action <code>lb_output</code>
> +        and avoids recirculation of packet in datapath. It is supported
> +        only for balance-tcp bond mode in netdev datapath. The new action
> +        gives higer performance but also requires pmd reloads whenever
> +        traffic load needs to be rebalanced among bonded links. By default
> +        this new action is disabled, however it can be enabled by setting
> +        <ref column="other-config" key="lb-output-action"/> in
> +        <ref table="Port"/> table.
> +      </column>
>        <group title="Connection-Tracking Capabilities">
>          <p>
>            These capabilities are granular because Open vSwitch and its
>
Ben Pfaff Jan. 21, 2020, 8:54 p.m. UTC | #6
On Mon, Jan 20, 2020 at 05:09:05PM +0100, Ilya Maximets wrote:
> > +static struct tx_bond *
> > +tx_bond_lookup(const struct hmap *hmap, uint32_t bond_id)
> > +{
> > +    struct tx_bond *tx;
> > +
> > +    HMAP_FOR_EACH_IN_BUCKET (tx, node, hash_bond_id(bond_id), hmap) {
> 
> Why not HMAP_FOR_EACH_WITH_HASH ?

HMAP_FOR_EACH_IN_BUCKET is a small optimization over
HMAP_FOR_EACH_WITH_HASH in cases where the comparison of the key is
cheap.  Comparing the hash is an optimization if the key is expensive to
compare (like an arbitrary string) but it just costs an extra integer
comparison if the key is a small integer as here.

> BTW, it's better to calculate hash before the loop to be sure that
> we're not recalculating it on each iteration.

A lot of code assumes that the hash only gets evaluated once.
Ilya Maximets Jan. 22, 2020, 7:49 p.m. UTC | #7
On 21.01.2020 21:54, Ben Pfaff wrote:
> On Mon, Jan 20, 2020 at 05:09:05PM +0100, Ilya Maximets wrote:
>>> +static struct tx_bond *
>>> +tx_bond_lookup(const struct hmap *hmap, uint32_t bond_id)
>>> +{
>>> +    struct tx_bond *tx;
>>> +
>>> +    HMAP_FOR_EACH_IN_BUCKET (tx, node, hash_bond_id(bond_id), hmap) {
>>
>> Why not HMAP_FOR_EACH_WITH_HASH ?
> 
> HMAP_FOR_EACH_IN_BUCKET is a small optimization over
> HMAP_FOR_EACH_WITH_HASH in cases where the comparison of the key is
> cheap.  Comparing the hash is an optimization if the key is expensive to
> compare (like an arbitrary string) but it just costs an extra integer
> comparison if the key is a small integer as here.

OK.  Thanks for explanation.
 
>> BTW, it's better to calculate hash before the loop to be sure that
>> we're not recalculating it on each iteration.
> 
> A lot of code assumes that the hash only gets evaluated once.
> 

Sure.  And there is no real issue here, but this just catches my eyes
for some reason.
Ilya Maximets Jan. 28, 2020, 7:09 p.m. UTC | #8
On 20.01.2020 17:09, Ilya Maximets wrote:
> On 08.01.2020 06:18, Vishal Deep Ajmera wrote:
>> Problem:
>> --------
>> In OVS-DPDK, flows with output over a bond interface of type “balance-tcp”
>> (using a hash on TCP/UDP 5-tuple) get translated by the ofproto layer into
>> "HASH" and "RECIRC" datapath actions. After recirculation, the packet is
>> forwarded to the bond member port based on 8-bits of the datapath hash
>> value computed through dp_hash. This causes performance degradation in the
>> following ways:
>>
>> 1. The recirculation of the packet implies another lookup of the packet’s
>> flow key in the exact match cache (EMC) and potentially Megaflow classifier
>> (DPCLS). This is the biggest cost factor.
>>
>> 2. The recirculated packets have a new “RSS” hash and compete with the
>> original packets for the scarce number of EMC slots. This implies more
>> EMC misses and potentially EMC thrashing causing costly DPCLS lookups.
>>
>> 3. The 256 extra megaflow entries per bond for dp_hash bond selection put
>> additional load on the revalidation threads.
>>
>> Owing to this performance degradation, deployments stick to “balance-slb”
>> bond mode even though it does not do active-active load balancing for
>> VXLAN- and GRE-tunnelled traffic because all tunnel packet have the same
>> source MAC address.
>>
>> Proposed optimization:
>> ----------------------
>> This proposal introduces a new load-balancing output action instead of
>> recirculation.
>>
>> Maintain one table per-bond (could just be an array of uint16's) and
>> program it the same way internal flows are created today for each possible
>> hash value(256 entries) from ofproto layer. Use this table to load-balance
>> flows as part of output action processing.
>>
>> Currently xlate_normal() -> output_normal() -> bond_update_post_recirc_rules()
>> -> bond_may_recirc() and compose_output_action__() generate
>> “dp_hash(hash_l4(0))” and “recirc(<RecircID>)” actions. In this case the
>> RecircID identifies the bond. For the recirculated packets the ofproto layer
>> installs megaflow entries that match on RecircID and masked dp_hash and send
>> them to the corresponding output port.
>>
>> Instead, we will now generate action as
>>     "lb_output(bond,<bond id>)"
>>
>> This combines hash computation (only if needed, else re-use RSS hash) and
>> inline load-balancing over the bond. This action is used *only* for balance-tcp
>> bonds in OVS-DPDK datapath (the OVS kernel datapath remains unchanged).
>>
>> Example:
>> --------
>> Current scheme:
>> ---------------
>> With 1 IP-UDP flow:
>>
>> flow-dump from pmd on cpu core: 2
>> recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2828969, bytes:181054016, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)
>>
>> recirc_id(0x1),dp_hash(0x113683bd/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:2828937, bytes:181051968, used:0.000s, actions:2
>>
>> With 8 IP-UDP flows (with random UDP src port): (all hitting same DPCL):
>>
>> flow-dump from pmd on cpu core: 2
>> recirc_id(0),in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:hash(hash_l4(0)),recirc(0x1)
>>
>> recirc_id(0x1),dp_hash(0xf8e02b7e/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:377395, bytes:24153280, used:0.000s, actions:2
>> recirc_id(0x1),dp_hash(0xb236c260/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:333486, bytes:21343104, used:0.000s, actions:1
>> recirc_id(0x1),dp_hash(0x7d89eb18/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:348461, bytes:22301504, used:0.000s, actions:1
>> recirc_id(0x1),dp_hash(0xa78d75df/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:633353, bytes:40534592, used:0.000s, actions:2
>> recirc_id(0x1),dp_hash(0xb58d846f/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:319901, bytes:20473664, used:0.001s, actions:2
>> recirc_id(0x1),dp_hash(0x24534406/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:334985, bytes:21439040, used:0.001s, actions:1
>> recirc_id(0x1),dp_hash(0x3cf32550/0xff),in_port(7),packet_type(ns=0,id=0),eth_type(0x0800),ipv4(frag=no), packets:326404, bytes:20889856, used:0.001s, actions:1
>>
>> New scheme:
>> -----------
>> We can do with a single flow entry (for any number of new flows):
>>
>> in_port(7),packet_type(ns=0,id=0),eth(src=02:00:00:02:14:01,dst=0c:c4:7a:58:f0:2b),eth_type(0x0800),ipv4(frag=no), packets:2674009, bytes:171136576, used:0.000s, actions:lb_output(bond,1)
>>
>> A new CLI has been added to dump datapath bond cache as given below.
>>
>> “sudo ovs-appctl dpif-netdev/dp-bond-show [dp]”
>>
>> root@ubuntu-190:performance_scripts # sudo ovs-appctl dpif-netdev/dp-bond-show
>> Bond cache:
>>         bond-id 1 :
>>                 bucket 0 - slave 2
>>                 bucket 1 - slave 1
>>                 bucket 2 - slave 2
>>                 bucket 3 - slave 1
>>
>> Performance improvement:
>> ------------------------
>> With a prototype of the proposed idea, the following perf improvement is seen
>> with Phy-VM-Phy UDP traffic, single flow. With multiple flows, the improvement
>> is even more enhanced (due to reduced number of flows).
>>
>> 1 VM:
>> *****
>> +--------------------------------------+
>> |                 mpps                 |
>> +--------------------------------------+
>> | Flows  master  with-opt.   %delta    |
>> +--------------------------------------+
>> | 1      4.47    5.73        28.21
>> | 10     4.17    5.35        28.45
>> | 400    3.52    5.39        53.01
>> | 1k     3.41    5.25        53.78
>> | 10k    2.53    4.57        80.44
>> | 100k   2.32    4.27        83.59
>> | 500k   2.32    4.27        83.59
>> +--------------------------------------+
>> mpps: million packets per second.
>> packet size 64 bytes.
>>
>> Signed-off-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
>> Co-authored-by: Manohar Krishnappa Chidambaraswamy <manukc@gmail.com>
>> Signed-off-by: Vishal Deep Ajmera <vishal.deep.ajmera@ericsson.com>
>>
>> CC: Jan Scheurich <jan.scheurich@ericsson.com>
>> CC: Venkatesan Pradeep <venkatesan.pradeep@ericsson.com>
>> CC: Nitin Katiyar <nitin.katiyar@ericsson.com>
>> ---
>>  datapath/linux/compat/include/linux/openvswitch.h |   1 +
>>  lib/dpif-netdev.c                                 | 502 ++++++++++++++++++++--
>>  lib/dpif-netlink.c                                |   3 +
>>  lib/dpif-provider.h                               |   9 +
>>  lib/dpif.c                                        |  49 +++
>>  lib/dpif.h                                        |   7 +
>>  lib/odp-execute.c                                 |   2 +
>>  lib/odp-util.c                                    |   4 +
>>  ofproto/bond.c                                    |  56 ++-
>>  ofproto/bond.h                                    |   9 +
>>  ofproto/ofproto-dpif-ipfix.c                      |   1 +
>>  ofproto/ofproto-dpif-sflow.c                      |   1 +
>>  ofproto/ofproto-dpif-xlate.c                      |  42 +-
>>  ofproto/ofproto-dpif.c                            |  24 ++
>>  ofproto/ofproto-dpif.h                            |   9 +-
>>  tests/lacp.at                                     |   9 +
>>  vswitchd/bridge.c                                 |   6 +
>>  vswitchd/vswitch.xml                              |  23 +
>>  18 files changed, 695 insertions(+), 62 deletions(-)
> 
> I started reviewing, but stopped at some point because I realized
> that current solution is over-complicated (too much of unnecessary
> memory copies) and not thread safe.
> See some comments inline.
> 
> I'll follow up with suggestions on how to implement this functionality
> in a different way, since I don't think that it's possible to make
> current version work correctly.

So, the root cause of all the issues in this patch, in my understanding,
is the fact that you need to collect statistics for all the bond hashes
in order to be able to rebalance traffic.  This forces you to have access
to PMD local caches.

The basic idea how to overcome this issue is to not have PMD local bond
cache, but have an RCU-protected data structure instead.

Memory layout scheme in current patch consists of 3 layers:
  1. Global hash map for all bonds. One for the whole datapath.
     Protected by dp->bond_mutex.
  2. Hash map of all the bonds. One for each PMD thread.
     Protected by pmd->bond_mutex.
  3. Hash map of all the bonds. Local copy that could be used
     lockless, but only by the owning PMD thread.

Suggested layout #1:
  Single global concurrent hash map (cmap) for all bonds. One for the whole
  datapath.  Bond addition/deletion protected by the dp->bond_mutex.
  Reads are lockless since protected by RCU.  Statistics updates must be
  fully atomic (i.e. atomic_add_relaxed()).

Suggested layout #2:
  One cmap for each PMD thread (no global one).  Bond addition/deletion
  protected by the pmd->bond_mutex.  Reads are lockless since protected
  by RCU.  Statistics updates should be atomic in terms of reads and writes.
  (non_atomic_ullong_add() function could be used).
  (This is similar to how we store per-PMD flow tables.)

#1 will consume a lot less memory, but could scale worse in case of too many
threads trying to send traffic to the same bond port.  #2 might be a bit
faster and more scalable in terms of performance, but less efficient in
memory consumption and might be slower in terms of response to slave updates
since we have to update hash maps on all the threads.
Both solutions doesn't require any reconfiguration of running PMD threads.

Note that to update cmap entry, you will likely need to prepare the new cmap
node and use a cmap_replace().  Statistics copy in this case might be a bit
tricky because you may lost part of additions within the period while PMD
threads are still using the old entry. To avoid that, statistics copying
should be RCU-postponed.  However, I'm not sure if we need highly accurate
stats there.

Any thoughts and suggestion (alternative solutions) are welcome to discuss.

Best regards, Ilya Maximets.
Jeff Squyres \(jsquyres\) via dev Jan. 30, 2020, 11:30 a.m. UTC | #9
>
> So, the root cause of all the issues in this patch, in my understanding,
> is the fact that you need to collect statistics for all the bond hashes
> in order to be able to rebalance traffic.  This forces you to have access
> to PMD local caches.
>
> The basic idea how to overcome this issue is to not have PMD local bond
> cache, but have an RCU-protected data structure instead.
>
> Memory layout scheme in current patch consists of 3 layers:
>   1. Global hash map for all bonds. One for the whole datapath.
>      Protected by dp->bond_mutex.
>   2. Hash map of all the bonds. One for each PMD thread.
>      Protected by pmd->bond_mutex.
>   3. Hash map of all the bonds. Local copy that could be used
>      lockless, but only by the owning PMD thread.
>
> Suggested layout #1:
>   Single global concurrent hash map (cmap) for all bonds. One for the whole
>   datapath.  Bond addition/deletion protected by the dp->bond_mutex.
>   Reads are lockless since protected by RCU.  Statistics updates must be
>   fully atomic (i.e. atomic_add_relaxed()).
>
> Suggested layout #2:
>   One cmap for each PMD thread (no global one).  Bond addition/deletion
>   protected by the pmd->bond_mutex.  Reads are lockless since protected
>   by RCU.  Statistics updates should be atomic in terms of reads and writes.
>   (non_atomic_ullong_add() function could be used).
>   (This is similar to how we store per-PMD flow tables.)
>
> #1 will consume a lot less memory, but could scale worse in case of too many
> threads trying to send traffic to the same bond port.  #2 might be a bit
> faster and more scalable in terms of performance, but less efficient in
> memory consumption and might be slower in terms of response to slave updates
> since we have to update hash maps on all the threads.
> Both solutions doesn't require any reconfiguration of running PMD threads.
>
> Note that to update cmap entry, you will likely need to prepare the new cmap
> node and use a cmap_replace().  Statistics copy in this case might be a bit
> tricky because you may lost part of additions within the period while PMD
> threads are still using the old entry. To avoid that, statistics copying
> should be RCU-postponed.  However, I'm not sure if we need highly accurate
> stats there.
>
> Any thoughts and suggestion (alternative solutions) are welcome to discuss.

Thanks Ilya. I will have a look at it and implement it in the next patch-set.

Warm Regards,
Vishal Ajmera
Jeff Squyres \(jsquyres\) via dev Feb. 6, 2020, 6:39 a.m. UTC | #10
> >
> > So, the root cause of all the issues in this patch, in my understanding,
> > is the fact that you need to collect statistics for all the bond hashes
> > in order to be able to rebalance traffic.  This forces you to have access
> > to PMD local caches.
> >
> > The basic idea how to overcome this issue is to not have PMD local bond
> > cache, but have an RCU-protected data structure instead.
> >
> > Memory layout scheme in current patch consists of 3 layers:
> >   1. Global hash map for all bonds. One for the whole datapath.
> >      Protected by dp->bond_mutex.
> >   2. Hash map of all the bonds. One for each PMD thread.
> >      Protected by pmd->bond_mutex.
> >   3. Hash map of all the bonds. Local copy that could be used
> >      lockless, but only by the owning PMD thread.
> >
> > Suggested layout #1:
> >   Single global concurrent hash map (cmap) for all bonds. One for the 
> > whole
> >   datapath.  Bond addition/deletion protected by the dp->bond_mutex.
> >   Reads are lockless since protected by RCU.  Statistics updates must be
> >   fully atomic (i.e. atomic_add_relaxed()).
> >
> > Suggested layout #2:
> >   One cmap for each PMD thread (no global one).  Bond addition/deletion
> >   protected by the pmd->bond_mutex.  Reads are lockless since protected
> >   by RCU.  Statistics updates should be atomic in terms of reads and 
> > writes.
> >   (non_atomic_ullong_add() function could be used).
> >   (This is similar to how we store per-PMD flow tables.)
> >
> > #1 will consume a lot less memory, but could scale worse in case of too 
> > many
> > threads trying to send traffic to the same bond port.  #2 might be a bit
> > faster and more scalable in terms of performance, but less efficient in
> > memory consumption and might be slower in terms of response to slave
> updates
> > since we have to update hash maps on all the threads.
> > Both solutions doesn't require any reconfiguration of running PMD threads.
> >
> > Note that to update cmap entry, you will likely need to prepare the new 
> > cmap
> > node and use a cmap_replace().  Statistics copy in this case might be a 
> > bit
> > tricky because you may lost part of additions within the period while PMD
> > threads are still using the old entry. To avoid that, statistics copying
> > should be RCU-postponed.  However, I'm not sure if we need highly accurate
> > stats there.
> >

Hi Ilya,

I am inclined to use option 2 to have per-pmd cmap of bonds however,
I see that even in this option we will need to have one cmap (or hmap)
of bonds at global datapath level. This will take care of 
reconfigure_pmd_threads
scenario. New PMDs needs to be configured with bond cmap from global map.

Warm Regards,
Vishal Ajmera
Ilya Maximets Feb. 6, 2020, 11 a.m. UTC | #11
On 2/6/20 7:39 AM, Vishal Deep Ajmera wrote:
>>>
>>> So, the root cause of all the issues in this patch, in my understanding,
>>> is the fact that you need to collect statistics for all the bond hashes
>>> in order to be able to rebalance traffic.  This forces you to have access
>>> to PMD local caches.
>>>
>>> The basic idea how to overcome this issue is to not have PMD local bond
>>> cache, but have an RCU-protected data structure instead.
>>>
>>> Memory layout scheme in current patch consists of 3 layers:
>>>   1. Global hash map for all bonds. One for the whole datapath.
>>>      Protected by dp->bond_mutex.
>>>   2. Hash map of all the bonds. One for each PMD thread.
>>>      Protected by pmd->bond_mutex.
>>>   3. Hash map of all the bonds. Local copy that could be used
>>>      lockless, but only by the owning PMD thread.
>>>
>>> Suggested layout #1:
>>>   Single global concurrent hash map (cmap) for all bonds. One for the 
>>> whole
>>>   datapath.  Bond addition/deletion protected by the dp->bond_mutex.
>>>   Reads are lockless since protected by RCU.  Statistics updates must be
>>>   fully atomic (i.e. atomic_add_relaxed()).
>>>
>>> Suggested layout #2:
>>>   One cmap for each PMD thread (no global one).  Bond addition/deletion
>>>   protected by the pmd->bond_mutex.  Reads are lockless since protected
>>>   by RCU.  Statistics updates should be atomic in terms of reads and 
>>> writes.
>>>   (non_atomic_ullong_add() function could be used).
>>>   (This is similar to how we store per-PMD flow tables.)
>>>
>>> #1 will consume a lot less memory, but could scale worse in case of too 
>>> many
>>> threads trying to send traffic to the same bond port.  #2 might be a bit
>>> faster and more scalable in terms of performance, but less efficient in
>>> memory consumption and might be slower in terms of response to slave
>> updates
>>> since we have to update hash maps on all the threads.
>>> Both solutions doesn't require any reconfiguration of running PMD threads.
>>>
>>> Note that to update cmap entry, you will likely need to prepare the new 
>>> cmap
>>> node and use a cmap_replace().  Statistics copy in this case might be a 
>>> bit
>>> tricky because you may lost part of additions within the period while PMD
>>> threads are still using the old entry. To avoid that, statistics copying
>>> should be RCU-postponed.  However, I'm not sure if we need highly accurate
>>> stats there.
>>>
> 
> Hi Ilya,
> 
> I am inclined to use option 2 to have per-pmd cmap of bonds however,
> I see that even in this option we will need to have one cmap (or hmap)
> of bonds at global datapath level. This will take care of 
> reconfigure_pmd_threads
> scenario. New PMDs needs to be configured with bond cmap from global map.

OK. Make sense for a newly created threads.
Jeff Squyres \(jsquyres\) via dev Feb. 12, 2020, 11:14 a.m. UTC | #12
> > I am inclined to use option 2 to have per-pmd cmap of bonds however,
> > I see that even in this option we will need to have one cmap (or hmap)
> > of bonds at global datapath level. This will take care of
> > reconfigure_pmd_threads
> > scenario. New PMDs needs to be configured with bond cmap from global
> map.
>
> OK. Make sense for a newly created threads.
>
Hi Ilya,

I have posted new patch addressing your comments. Let me know if it looks ok 
now.

Warm Regards,
Vishal Ajmera

Patch
diff mbox series

diff --git a/datapath/linux/compat/include/linux/openvswitch.h b/datapath/linux/compat/include/linux/openvswitch.h
index 2f0c655..7604a2a 100644
--- a/datapath/linux/compat/include/linux/openvswitch.h
+++ b/datapath/linux/compat/include/linux/openvswitch.h
@@ -1021,6 +1021,7 @@  enum ovs_action_attr {
 	OVS_ACTION_ATTR_TUNNEL_PUSH,   /* struct ovs_action_push_tnl*/
 	OVS_ACTION_ATTR_TUNNEL_POP,    /* u32 port number. */
 	OVS_ACTION_ATTR_DROP,          /* u32 xlate_error. */
+	OVS_ACTION_ATTR_LB_OUTPUT,     /* u32 bond-id */
 #endif
 	__OVS_ACTION_ATTR_MAX,	      /* Nothing past this will be accepted
 				       * from userspace. */
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 2421821..78f9cf9 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -79,6 +79,7 @@ 
 #include "unixctl.h"
 #include "util.h"
 #include "uuid.h"
+#include "ofproto/bond.h"
 
 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
 
@@ -377,6 +378,11 @@  struct dp_netdev {
 
     struct conntrack *conntrack;
     struct pmd_auto_lb pmd_alb;
+    /* Bonds.
+     *
+     * Any lookup into 'bonds' requires taking 'bond_mutex'. */
+    struct ovs_mutex bond_mutex;
+    struct hmap bonds;
 };
 
 static void meter_lock(const struct dp_netdev *dp, uint32_t meter_id)
@@ -607,6 +613,20 @@  struct tx_port {
     struct dp_netdev_rxq *output_pkts_rxqs[NETDEV_MAX_BURST];
 };
 
+/* Contained by struct tx_bond 'slave_buckets' */
+struct slave_entry {
+    odp_port_t slave_id;
+    atomic_ullong n_packets;
+    atomic_ullong n_bytes;
+};
+
+/* Contained by struct dp_netdev_pmd_thread's 'bond_cache' or 'tx_bonds'. */
+struct tx_bond {
+    struct hmap_node node;
+    uint32_t bond_id;
+    struct slave_entry slave_buckets[BOND_BUCKETS];
+};
+
 /* A set of properties for the current processing loop that is not directly
  * associated with the pmd thread itself, but with the packets being
  * processed or the short-term system configuration (for example, time).
@@ -718,7 +738,6 @@  struct dp_netdev_pmd_thread {
     atomic_bool wait_for_reload;    /* Can we busy wait for the next reload? */
     atomic_bool reload_tx_qid;      /* Do we need to reload static_tx_qid? */
     atomic_bool exit;               /* For terminating the pmd thread. */
-
     pthread_t thread;
     unsigned core_id;               /* CPU core id of this pmd thread. */
     int numa_id;                    /* numa node id of this pmd thread. */
@@ -739,6 +758,11 @@  struct dp_netdev_pmd_thread {
      * read by the pmd thread. */
     struct hmap tx_ports OVS_GUARDED;
 
+    struct ovs_mutex bond_mutex;    /* Mutex for 'tx_bonds'. */
+    /* Map of 'tx_bond's used for transmission.  Written by the main thread,
+     * read/written by the pmd thread. */
+    struct hmap tx_bonds OVS_GUARDED;
+
     /* These are thread-local copies of 'tx_ports'.  One contains only tunnel
      * ports (that support push_tunnel/pop_tunnel), the other contains ports
      * with at least one txq (that support send).  A port can be in both.
@@ -751,6 +775,8 @@  struct dp_netdev_pmd_thread {
      * other instance will only be accessed by its own pmd thread. */
     struct hmap tnl_port_cache;
     struct hmap send_port_cache;
+    /* These are thread-local copies of 'tx_bonds' */
+    struct hmap bond_cache;
 
     /* Keep track of detailed PMD performance statistics. */
     struct pmd_perf_stats perf_stats;
@@ -830,6 +856,12 @@  static void dp_netdev_del_rxq_from_pmd(struct dp_netdev_pmd_thread *pmd,
 static int
 dp_netdev_pmd_flush_output_packets(struct dp_netdev_pmd_thread *pmd,
                                    bool force);
+static void dp_netdev_add_bond_tx_to_pmd(struct dp_netdev_pmd_thread *pmd,
+                                         struct tx_bond *bond)
+    OVS_REQUIRES(pmd->bond_mutex);
+static void dp_netdev_del_bond_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
+                                           struct tx_bond *tx)
+    OVS_REQUIRES(pmd->bond_mutex);
 
 static void reconfigure_datapath(struct dp_netdev *dp)
     OVS_REQUIRES(dp->port_mutex);
@@ -838,6 +870,10 @@  static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd);
 static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd);
 static void pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd)
     OVS_REQUIRES(pmd->port_mutex);
+static void pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
+    OVS_REQUIRES(pmd->bond_mutex);
+static void pmd_load_bond_cache(struct dp_netdev_pmd_thread *pmd);
+
 static inline void
 dp_netdev_pmd_try_optimize(struct dp_netdev_pmd_thread *pmd,
                            struct polled_queue *poll_list, int poll_cnt);
@@ -1396,6 +1432,45 @@  pmd_perf_show_cmd(struct unixctl_conn *conn, int argc,
     par.command_type = PMD_INFO_PERF_SHOW;
     dpif_netdev_pmd_info(conn, argc, argv, &par);
 }
+
+static void
+dpif_netdev_dp_bond_show(struct unixctl_conn *conn, int argc,
+                         const char *argv[], void *aux OVS_UNUSED)
+{
+    struct ds reply = DS_EMPTY_INITIALIZER;
+
+    ovs_mutex_lock(&dp_netdev_mutex);
+
+    struct dp_netdev *dp = NULL;
+    if (argc == 2) {
+        dp = shash_find_data(&dp_netdevs, argv[1]);
+    } else if (shash_count(&dp_netdevs) == 1) {
+        /* There's only one datapath */
+        dp = shash_first(&dp_netdevs)->data;
+    }
+    if (!dp) {
+        ovs_mutex_unlock(&dp_netdev_mutex);
+        unixctl_command_reply_error(conn,
+                                    "please specify an existing datapath");
+        return;
+    }
+    ds_put_cstr(&reply, "\nBonds:\n");
+
+    struct tx_bond *dp_bond_entry;
+    HMAP_FOR_EACH (dp_bond_entry, node, &dp->bonds) {
+        ds_put_format(&reply, "\tbond-id %u :\n",
+                      dp_bond_entry->bond_id);
+        for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
+            ds_put_format(&reply, "\t\tbucket %u - slave %u \n",
+                      bucket,
+                      dp_bond_entry->slave_buckets[bucket].slave_id);
+        }
+    }
+    ovs_mutex_unlock(&dp_netdev_mutex);
+    unixctl_command_reply(conn, ds_cstr(&reply));
+    ds_destroy(&reply);
+}
+
 
 static int
 dpif_netdev_init(void)
@@ -1427,6 +1502,9 @@  dpif_netdev_init(void)
                              "[-us usec] [-q qlen]",
                              0, 10, pmd_perf_log_set_cmd,
                              NULL);
+    unixctl_command_register("dpif-netdev/dp-bond-show", "[dp]",
+                             0, 1, dpif_netdev_dp_bond_show,
+                             NULL);
     return 0;
 }
 
@@ -1551,6 +1629,9 @@  create_dp_netdev(const char *name, const struct dpif_class *class,
     ovs_mutex_init(&dp->port_mutex);
     hmap_init(&dp->ports);
     dp->port_seq = seq_create();
+    ovs_mutex_init(&dp->bond_mutex);
+    hmap_init(&dp->bonds);
+
     fat_rwlock_init(&dp->upcall_rwlock);
 
     dp->reconfigure_seq = seq_create();
@@ -1674,6 +1755,14 @@  dp_netdev_free(struct dp_netdev *dp)
     }
     ovs_mutex_unlock(&dp->port_mutex);
 
+    ovs_mutex_lock(&dp->bond_mutex);
+    struct tx_bond *bond, *next_bond;
+    HMAP_FOR_EACH_SAFE (bond, next_bond, node, &dp->bonds) {
+        hmap_remove(&dp->bonds, &bond->node);
+        free(bond);
+    }
+    ovs_mutex_unlock(&dp->bond_mutex);
+
     dp_netdev_destroy_all_pmds(dp, true);
     cmap_destroy(&dp->poll_threads);
 
@@ -1692,6 +1781,9 @@  dp_netdev_free(struct dp_netdev *dp)
     hmap_destroy(&dp->ports);
     ovs_mutex_destroy(&dp->port_mutex);
 
+    hmap_destroy(&dp->bonds);
+    ovs_mutex_destroy(&dp->bond_mutex);
+
     /* Upcalls must be disabled at this point */
     dp_netdev_destroy_upcall_lock(dp);
 
@@ -1795,6 +1887,7 @@  dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd)
         ovs_mutex_lock(&pmd->port_mutex);
         pmd_load_cached_ports(pmd);
         ovs_mutex_unlock(&pmd->port_mutex);
+        pmd_load_bond_cache(pmd);
         ovs_mutex_unlock(&pmd->dp->non_pmd_mutex);
         return;
     }
@@ -1809,6 +1902,12 @@  hash_port_no(odp_port_t port_no)
     return hash_int(odp_to_u32(port_no), 0);
 }
 
+static uint32_t
+hash_bond_id(uint32_t bond_id)
+{
+    return hash_int(bond_id, 0);
+}
+
 static int
 port_create(const char *devname, const char *type,
             odp_port_t port_no, struct dp_netdev_port **portp)
@@ -4349,6 +4448,19 @@  tx_port_lookup(const struct hmap *hmap, odp_port_t port_no)
     return NULL;
 }
 
+static struct tx_bond *
+tx_bond_lookup(const struct hmap *hmap, uint32_t bond_id)
+{
+    struct tx_bond *tx;
+
+    HMAP_FOR_EACH_IN_BUCKET (tx, node, hash_bond_id(bond_id), hmap) {
+        if (tx->bond_id == bond_id) {
+            return tx;
+        }
+    }
+    return NULL;
+}
+
 static int
 port_reconfigure(struct dp_netdev_port *port)
 {
@@ -4837,6 +4949,26 @@  pmd_remove_stale_ports(struct dp_netdev *dp,
     ovs_mutex_unlock(&pmd->port_mutex);
 }
 
+static void
+pmd_remove_stale_bonds(struct dp_netdev *dp,
+                       struct dp_netdev_pmd_thread *pmd)
+    OVS_EXCLUDED(pmd->bond_mutex)
+    OVS_EXCLUDED(dp->bond_mutex)
+{
+    ovs_mutex_lock(&dp->bond_mutex);
+    ovs_mutex_lock(&pmd->bond_mutex);
+
+    struct tx_bond *tx, *tx_next;
+    HMAP_FOR_EACH_SAFE (tx, tx_next, node, &pmd->tx_bonds) {
+        if (!tx_bond_lookup(&dp->bonds, tx->bond_id)) {
+            dp_netdev_del_bond_tx_from_pmd(pmd, tx);
+        }
+    }
+
+    ovs_mutex_unlock(&pmd->bond_mutex);
+    ovs_mutex_unlock(&dp->bond_mutex);
+}
+
 /* Must be called each time a port is added/removed or the cmask changes.
  * This creates and destroys pmd threads, reconfigures ports, opens their
  * rxqs and assigns all rxqs/txqs to pmd threads. */
@@ -4875,10 +5007,11 @@  reconfigure_datapath(struct dp_netdev *dp)
         }
     }
 
-    /* Remove from the pmd threads all the ports that have been deleted or
-     * need reconfiguration. */
+    /* Remove from the pmd threads all the ports/bonds that have been deleted
+     * or need reconfiguration. */
     CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
         pmd_remove_stale_ports(dp, pmd);
+        pmd_remove_stale_bonds(dp, pmd);
     }
 
     /* Reload affected pmd threads.  We must wait for the pmd threads before
@@ -5000,6 +5133,21 @@  reconfigure_datapath(struct dp_netdev *dp)
         ovs_mutex_unlock(&pmd->port_mutex);
     }
 
+    /* Add every bond to the tx cache of every pmd thread, if it's not
+     * there already and if this pmd has at least one rxq to poll. */
+    ovs_mutex_lock(&dp->bond_mutex);
+    CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
+        ovs_mutex_lock(&pmd->bond_mutex);
+        if (hmap_count(&pmd->poll_list) || pmd->core_id == NON_PMD_CORE_ID) {
+            struct tx_bond *bond;
+            HMAP_FOR_EACH (bond, node, &dp->bonds) {
+                dp_netdev_add_bond_tx_to_pmd(pmd, bond);
+            }
+        }
+        ovs_mutex_unlock(&pmd->bond_mutex);
+    }
+    ovs_mutex_unlock(&dp->bond_mutex);
+
     /* Reload affected pmd threads. */
     reload_affected_pmds(dp);
 
@@ -5258,7 +5406,6 @@  pmd_rebalance_dry_run(struct dp_netdev *dp)
     return ret;
 }
 
-
 /* Return true if needs to revalidate datapath flows. */
 static bool
 dpif_netdev_run(struct dpif *dpif)
@@ -5428,6 +5575,56 @@  pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd)
 }
 
 static void
+pmd_free_cached_bonds(struct dp_netdev_pmd_thread *pmd)
+{
+    struct tx_bond *bond, *next;
+
+    /* Remove bonds from pmd which no longer exists. */
+    HMAP_FOR_EACH_SAFE (bond, next, node, &pmd->bond_cache) {
+        struct tx_bond *tx = NULL;
+
+        tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);
+        if (!tx) {
+            /* Bond no longer exist. Delete it from pmd. */
+            hmap_remove(&pmd->bond_cache, &bond->node);
+            free(bond);
+        }
+    }
+}
+
+/* Copies bonds from 'pmd->tx_bonds' (shared with the main thread) to
+ * 'pmd->bond_cache' (thread local) */
+static void
+pmd_load_cached_bonds(struct dp_netdev_pmd_thread *pmd)
+    OVS_REQUIRES(pmd->bond_mutex)
+{
+    pmd_free_cached_bonds(pmd);
+    hmap_shrink(&pmd->bond_cache);
+
+    struct tx_bond *tx_bond, *tx_bond_cached;
+    HMAP_FOR_EACH (tx_bond, node, &pmd->tx_bonds) {
+        /* Check if bond already exist on pmd. */
+        tx_bond_cached = tx_bond_lookup(&pmd->bond_cache, tx_bond->bond_id);
+
+        if (!tx_bond_cached) {
+            /* Create new bond entry in cache. */
+            tx_bond_cached = xmemdup(tx_bond, sizeof *tx_bond_cached);
+            hmap_insert(&pmd->bond_cache, &tx_bond_cached->node,
+                        hash_bond_id(tx_bond_cached->bond_id));
+        } else {
+            /* Update the slave-map. */
+            for (int bucket = 0; bucket <= BOND_MASK; bucket++) {
+                tx_bond_cached->slave_buckets[bucket].slave_id =
+                    tx_bond->slave_buckets[bucket].slave_id;
+            }
+        }
+        VLOG_DBG("Caching bond-id %d pmd %d",
+                 tx_bond_cached->bond_id, pmd->core_id);
+    }
+}
+
+
+static void
 pmd_alloc_static_tx_qid(struct dp_netdev_pmd_thread *pmd)
 {
     ovs_mutex_lock(&pmd->dp->tx_qid_pool_mutex);
@@ -5449,6 +5646,14 @@  pmd_free_static_tx_qid(struct dp_netdev_pmd_thread *pmd)
     ovs_mutex_unlock(&pmd->dp->tx_qid_pool_mutex);
 }
 
+static void
+pmd_load_bond_cache(struct dp_netdev_pmd_thread *pmd)
+{
+    ovs_mutex_lock(&pmd->bond_mutex);
+    pmd_load_cached_bonds(pmd);
+    ovs_mutex_unlock(&pmd->bond_mutex);
+}
+
 static int
 pmd_load_queues_and_ports(struct dp_netdev_pmd_thread *pmd,
                           struct polled_queue **ppoll_list)
@@ -5476,6 +5681,8 @@  pmd_load_queues_and_ports(struct dp_netdev_pmd_thread *pmd,
 
     ovs_mutex_unlock(&pmd->port_mutex);
 
+    pmd_load_bond_cache(pmd);
+
     *ppoll_list = poll_list;
     return i;
 }
@@ -6038,6 +6245,7 @@  dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp,
     atomic_init(&pmd->reload, false);
     ovs_mutex_init(&pmd->flow_mutex);
     ovs_mutex_init(&pmd->port_mutex);
+    ovs_mutex_init(&pmd->bond_mutex);
     cmap_init(&pmd->flow_table);
     cmap_init(&pmd->classifiers);
     pmd->ctx.last_rxq = NULL;
@@ -6048,6 +6256,8 @@  dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp,
     hmap_init(&pmd->tx_ports);
     hmap_init(&pmd->tnl_port_cache);
     hmap_init(&pmd->send_port_cache);
+    hmap_init(&pmd->tx_bonds);
+    hmap_init(&pmd->bond_cache);
     /* init the 'flow_cache' since there is no
      * actual thread created for NON_PMD_CORE_ID. */
     if (core_id == NON_PMD_CORE_ID) {
@@ -6068,6 +6278,8 @@  dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd)
     hmap_destroy(&pmd->send_port_cache);
     hmap_destroy(&pmd->tnl_port_cache);
     hmap_destroy(&pmd->tx_ports);
+    hmap_destroy(&pmd->bond_cache);
+    hmap_destroy(&pmd->tx_bonds);
     hmap_destroy(&pmd->poll_list);
     /* All flows (including their dpcls_rules) have been deleted already */
     CMAP_FOR_EACH (cls, node, &pmd->classifiers) {
@@ -6079,6 +6291,7 @@  dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd)
     ovs_mutex_destroy(&pmd->flow_mutex);
     seq_destroy(pmd->reload_seq);
     ovs_mutex_destroy(&pmd->port_mutex);
+    ovs_mutex_destroy(&pmd->bond_mutex);
     free(pmd);
 }
 
@@ -6232,6 +6445,45 @@  dp_netdev_del_port_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
     free(tx);
     pmd->need_reload = true;
 }
+
+static void
+dp_netdev_add_bond_tx_to_pmd(struct dp_netdev_pmd_thread *pmd,
+                             struct tx_bond *bond)
+    OVS_REQUIRES(pmd->bond_mutex)
+{
+    bool reload = false;
+    struct tx_bond *tx = tx_bond_lookup(&pmd->tx_bonds, bond->bond_id);
+    if (tx) {
+        /* Check if mapping is changed. */
+        for (int i = 0; i <= BOND_MASK; i++) {
+            if (bond->slave_buckets[i].slave_id !=
+                     tx->slave_buckets[i].slave_id) {
+                /* Mapping is modified. Reload pmd bond cache again. */
+                reload = true;
+            }
+            /* Copy the map always. */
+            tx->slave_buckets[i].slave_id = bond->slave_buckets[i].slave_id;
+        }
+    } else {
+        tx = xmemdup(bond, sizeof *tx);
+        hmap_insert(&pmd->tx_bonds, &tx->node, hash_bond_id(bond->bond_id));
+        reload = true;
+    }
+    if (reload) {
+        pmd->need_reload = true;
+    }
+}
+
+/* Del 'tx' from the tx bond cache of 'pmd' */
+static void
+dp_netdev_del_bond_tx_from_pmd(struct dp_netdev_pmd_thread *pmd,
+                               struct tx_bond *tx)
+    OVS_REQUIRES(pmd->bond_mutex)
+{
+    hmap_remove(&pmd->tx_bonds, &tx->node);
+    free(tx);
+    pmd->need_reload = true;
+}
 
 static char *
 dpif_netdev_get_datapath_version(void)
@@ -7007,6 +7259,13 @@  pmd_send_port_cache_lookup(const struct dp_netdev_pmd_thread *pmd,
     return tx_port_lookup(&pmd->send_port_cache, port_no);
 }
 
+static struct tx_bond *
+pmd_tx_bond_cache_lookup(const struct dp_netdev_pmd_thread *pmd,
+                         uint32_t bond_id)
+{
+    return tx_bond_lookup(&pmd->bond_cache, bond_id);
+}
+
 static int
 push_tnl_action(const struct dp_netdev_pmd_thread *pmd,
                 const struct nlattr *attr,
@@ -7057,6 +7316,91 @@  dp_execute_userspace_action(struct dp_netdev_pmd_thread *pmd,
     }
 }
 
+static bool
+dp_execute_output_action(struct dp_netdev_pmd_thread *pmd,
+                         struct dp_packet_batch *packets_,
+                         bool should_steal, odp_port_t port_no)
+{
+    struct tx_port *p = pmd_send_port_cache_lookup(pmd, port_no);
+    if (!OVS_LIKELY(p)) {
+        return false;
+    }
+
+    struct dp_packet_batch out;
+    if (!should_steal) {
+        dp_packet_batch_clone(&out, packets_);
+        dp_packet_batch_reset_cutlen(packets_);
+        packets_ = &out;
+    }
+    dp_packet_batch_apply_cutlen(packets_);
+#ifdef DPDK_NETDEV
+    if (OVS_UNLIKELY(!dp_packet_batch_is_empty(&p->output_pkts)
+                     && packets_->packets[0]->source
+                     != p->output_pkts.packets[0]->source)) {
+        /* netdev-dpdk assumes that all packets in a single
+         * output batch has the same source. Flush here to
+         * avoid memory access issues. */
+        dp_netdev_pmd_flush_output_on_port(pmd, p);
+    }
+#endif
+    if (dp_packet_batch_size(&p->output_pkts)
+        + dp_packet_batch_size(packets_) > NETDEV_MAX_BURST) {
+        /* Flush here to avoid overflow. */
+        dp_netdev_pmd_flush_output_on_port(pmd, p);
+    }
+    if (dp_packet_batch_is_empty(&p->output_pkts)) {
+        pmd->n_output_batches++;
+    }
+
+    struct dp_packet *packet;
+    DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
+        p->output_pkts_rxqs[dp_packet_batch_size(&p->output_pkts)] =
+            pmd->ctx.last_rxq;
+        dp_packet_batch_add(&p->output_pkts, packet);
+    }
+    return true;
+}
+
+static bool
+dp_execute_lb_output_action(struct dp_netdev_pmd_thread *pmd,
+                            struct dp_packet_batch *packets_,
+                            bool should_steal, uint32_t bond)
+{
+    struct tx_bond *p_bond = pmd_tx_bond_cache_lookup(pmd, bond);
+    if (!p_bond) {
+        return false;
+    }
+
+    struct dp_packet_batch del_pkts;
+    dp_packet_batch_init(&del_pkts);
+
+    struct dp_packet *packet;
+    DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
+        /*
+         * Lookup the bond-hash table using hash to get the slave.
+         */
+        uint32_t hash = dp_packet_get_rss_hash(packet);
+        struct slave_entry *s_entry = &p_bond->slave_buckets[hash & BOND_MASK];
+        odp_port_t bond_member = s_entry->slave_id;
+        uint32_t size = dp_packet_size(packet);
+
+        struct dp_packet_batch output_pkt;
+        dp_packet_batch_init_packet(&output_pkt, packet);
+        if (OVS_LIKELY(dp_execute_output_action(pmd, &output_pkt, should_steal,
+                                                bond_member))) {
+            /* Update slave stats. */
+            non_atomic_ullong_add(&s_entry->n_packets, 1);
+            non_atomic_ullong_add(&s_entry->n_bytes, size);
+        } else {
+            dp_packet_batch_add(&del_pkts, packet);
+        }
+    }
+
+    /* Delete packets that failed OUTPUT action */
+    dp_packet_delete_batch(&del_pkts, should_steal);
+    return true;
+}
+
 static void
 dp_execute_cb(void *aux_, struct dp_packet_batch *packets_,
               const struct nlattr *a, bool should_steal)
@@ -7072,43 +7416,15 @@  dp_execute_cb(void *aux_, struct dp_packet_batch *packets_,
 
     switch ((enum ovs_action_attr)type) {
     case OVS_ACTION_ATTR_OUTPUT:
-        p = pmd_send_port_cache_lookup(pmd, nl_attr_get_odp_port(a));
-        if (OVS_LIKELY(p)) {
-            struct dp_packet *packet;
-            struct dp_packet_batch out;
-
-            if (!should_steal) {
-                dp_packet_batch_clone(&out, packets_);
-                dp_packet_batch_reset_cutlen(packets_);
-                packets_ = &out;
-            }
-            dp_packet_batch_apply_cutlen(packets_);
-
-#ifdef DPDK_NETDEV
-            if (OVS_UNLIKELY(!dp_packet_batch_is_empty(&p->output_pkts)
-                             && packets_->packets[0]->source
-                                != p->output_pkts.packets[0]->source)) {
-                /* XXX: netdev-dpdk assumes that all packets in a single
-                 *      output batch has the same source. Flush here to
-                 *      avoid memory access issues. */
-                dp_netdev_pmd_flush_output_on_port(pmd, p);
-            }
-#endif
-            if (dp_packet_batch_size(&p->output_pkts)
-                + dp_packet_batch_size(packets_) > NETDEV_MAX_BURST) {
-                /* Flush here to avoid overflow. */
-                dp_netdev_pmd_flush_output_on_port(pmd, p);
-            }
-
-            if (dp_packet_batch_is_empty(&p->output_pkts)) {
-                pmd->n_output_batches++;
-            }
+        if (dp_execute_output_action(pmd, packets_, should_steal,
+                                     nl_attr_get_odp_port(a))) {
+            return;
+        }
+        break;
 
-            DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
-                p->output_pkts_rxqs[dp_packet_batch_size(&p->output_pkts)] =
-                                                             pmd->ctx.last_rxq;
-                dp_packet_batch_add(&p->output_pkts, packet);
-            }
+    case OVS_ACTION_ATTR_LB_OUTPUT:
+        if (dp_execute_lb_output_action(pmd, packets_, should_steal,
+                                        nl_attr_get_u32(a))) {
             return;
         } else {
             COVERAGE_ADD(datapath_drop_invalid_port,
@@ -7666,6 +7982,109 @@  dpif_netdev_ipf_dump_done(struct dpif *dpif OVS_UNUSED, void *ipf_dump_ctx)
 
 }
 
+static int
+dpif_netdev_bond_add(struct dpif *dpif, uint32_t bond_id,
+                     odp_port_t slave_map[])
+{
+    struct dp_netdev *dp = get_dp_netdev(dpif);
+
+    ovs_mutex_lock(&dp->bond_mutex);
+    /*
+     * Lookup for the bond. If already exists, just update the slave-map.
+     * Else create new.
+     */
+    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
+    if (dp_bond_entry) {
+        for (int bucket = 0; bucket <= BOND_MASK; bucket++) {
+            dp_bond_entry->slave_buckets[bucket].slave_id = slave_map[bucket];
+        }
+
+        struct dp_netdev_pmd_thread *pmd;
+        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
+            ovs_mutex_lock(&pmd->bond_mutex);
+            dp_netdev_add_bond_tx_to_pmd(pmd, dp_bond_entry);
+            ovs_mutex_unlock(&pmd->bond_mutex);
+        }
+    } else {
+        struct tx_bond *dp_bond = xzalloc(sizeof *dp_bond);
+        dp_bond->bond_id = bond_id;
+        for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
+            dp_bond->slave_buckets[bucket].slave_id = slave_map[bucket];
+        }
+        hmap_insert(&dp->bonds, &dp_bond->node,
+                    hash_bond_id(dp_bond->bond_id));
+
+        /* Insert the bond map in all pmds. */
+        struct dp_netdev_pmd_thread *pmd;
+        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
+            ovs_mutex_lock(&pmd->bond_mutex);
+            dp_netdev_add_bond_tx_to_pmd(pmd, dp_bond);
+            ovs_mutex_unlock(&pmd->bond_mutex);
+        }
+    }
+    ovs_mutex_unlock(&dp->bond_mutex);
+    return 0;
+}
+
+static int
+dpif_netdev_bond_del(struct dpif *dpif, uint32_t bond_id)
+{
+    struct dp_netdev *dp = get_dp_netdev(dpif);
+
+    ovs_mutex_lock(&dp->bond_mutex);
+
+    /* Find the bond and delete it if present */
+    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
+    if (dp_bond_entry) {
+        /* Remove the bond map in all pmds. */
+        struct dp_netdev_pmd_thread *pmd;
+        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
+            ovs_mutex_lock(&pmd->bond_mutex);
+            dp_netdev_del_bond_tx_from_pmd(pmd, dp_bond_entry);
+            ovs_mutex_unlock(&pmd->bond_mutex);
+        }
+        hmap_remove(&dp->bonds, &dp_bond_entry->node);
+        free(dp_bond_entry);
+    }
+
+    ovs_mutex_unlock(&dp->bond_mutex);
+    return 0;
+}
+
+static int
+dpif_netdev_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
+                           uint64_t *n_bytes)
+{
+    struct dp_netdev *dp = get_dp_netdev(dpif);
+
+    ovs_mutex_lock(&dp->bond_mutex);
+
+    /* Find the bond and retrieve stats if present */
+    struct tx_bond *dp_bond_entry = tx_bond_lookup(&dp->bonds, bond_id);
+    if (dp_bond_entry) {
+        /* Search the bond in all PMDs */
+        struct dp_netdev_pmd_thread *pmd;
+        CMAP_FOR_EACH (pmd, node, &dp->poll_threads) {
+            ovs_mutex_lock(&pmd->bond_mutex);
+            struct tx_bond *pmd_bond_entry
+                = tx_bond_lookup(&pmd->bond_cache, bond_id);
+            if (pmd_bond_entry) {
+                /* Read bond stats. */
+                for (int i = 0; i <= BOND_MASK; i++) {
+                    uint64_t pmd_n_bytes;
+                    atomic_read_relaxed(
+                         &pmd_bond_entry->slave_buckets[i].n_bytes,
+                         &pmd_n_bytes);
+                    n_bytes[i] += pmd_n_bytes;
+                }
+            }
+            ovs_mutex_unlock(&pmd->bond_mutex);
+        }
+    }
+    ovs_mutex_unlock(&dp->bond_mutex);
+    return 0;
+}
+
 const struct dpif_class dpif_netdev_class = {
     "netdev",
     true,                       /* cleanup_required */
@@ -7739,6 +8158,9 @@  const struct dpif_class dpif_netdev_class = {
     dpif_netdev_meter_set,
     dpif_netdev_meter_get,
     dpif_netdev_meter_del,
+    dpif_netdev_bond_add,
+    dpif_netdev_bond_del,
+    dpif_netdev_bond_stats_get,
 };
 
 static void
diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
index 3d97c78..b1555df 100644
--- a/lib/dpif-netlink.c
+++ b/lib/dpif-netlink.c
@@ -3995,6 +3995,9 @@  const struct dpif_class dpif_netlink_class = {
     dpif_netlink_meter_set,
     dpif_netlink_meter_get,
     dpif_netlink_meter_del,
+    NULL,                       /* bond_add */
+    NULL,                       /* bond_del */
+    NULL,                       /* bond_stats_get */
 };
 
 static int
diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h
index b77317b..4415836 100644
--- a/lib/dpif-provider.h
+++ b/lib/dpif-provider.h
@@ -616,6 +616,15 @@  struct dpif_class {
      * zero. */
     int (*meter_del)(struct dpif *, ofproto_meter_id meter_id,
                      struct ofputil_meter_stats *, uint16_t n_bands);
+
+    /* Adds a bond with 'bond_id' and the slave-map to 'dpif'. */
+    int (*bond_add)(struct dpif *dpif, uint32_t bond_id,
+                    odp_port_t slave_map[]);
+    /* Removes bond identified by 'bond_id' from 'dpif'. */
+    int (*bond_del)(struct dpif *dpif, uint32_t bond_id);
+    /* Reads bond stats from 'dpif'. */
+    int (*bond_stats_get)(struct dpif *dpif, uint32_t bond_id,
+                          uint64_t *n_bytes);
 };
 
 extern const struct dpif_class dpif_netlink_class;
diff --git a/lib/dpif.c b/lib/dpif.c
index 955eaf6..41c0886 100644
--- a/lib/dpif.c
+++ b/lib/dpif.c
@@ -1186,6 +1186,7 @@  dpif_execute_helper_cb(void *aux_, struct dp_packet_batch *packets_,
 
     case OVS_ACTION_ATTR_CT:
     case OVS_ACTION_ATTR_OUTPUT:
+    case OVS_ACTION_ATTR_LB_OUTPUT:
     case OVS_ACTION_ATTR_TUNNEL_PUSH:
     case OVS_ACTION_ATTR_TUNNEL_POP:
     case OVS_ACTION_ATTR_USERSPACE:
@@ -1236,6 +1237,7 @@  dpif_execute_helper_cb(void *aux_, struct dp_packet_batch *packets_,
         struct dp_packet *clone = NULL;
         uint32_t cutlen = dp_packet_get_cutlen(packet);
         if (cutlen && (type == OVS_ACTION_ATTR_OUTPUT
+                        || type == OVS_ACTION_ATTR_LB_OUTPUT
                         || type == OVS_ACTION_ATTR_TUNNEL_PUSH
                         || type == OVS_ACTION_ATTR_TUNNEL_POP
                         || type == OVS_ACTION_ATTR_USERSPACE)) {
@@ -1895,6 +1897,16 @@  dpif_supports_explicit_drop_action(const struct dpif *dpif)
     return dpif_is_netdev(dpif);
 }
 
+bool
+dpif_supports_lb_output_action(const struct dpif *dpif)
+{
+    /*
+     * Balance-tcp optimization is currently supported in netdev
+     * datapath only.
+     */
+    return dpif_is_netdev(dpif);
+}
+
 /* Meters */
 void
 dpif_meter_get_features(const struct dpif *dpif,
@@ -1992,3 +2004,40 @@  dpif_meter_del(struct dpif *dpif, ofproto_meter_id meter_id,
     }
     return error;
 }
+
+int
+dpif_bond_add(struct dpif *dpif, uint32_t bond_id, odp_port_t slave_map[])
+{
+    int error = -1;
+
+    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_del) {
+        error = dpif->dpif_class->bond_add(dpif, bond_id, slave_map);
+    }
+
+    return error;
+}
+
+int
+dpif_bond_del(struct dpif *dpif, uint32_t bond_id)
+{
+    int error = -1;
+
+    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_del) {
+        error = dpif->dpif_class->bond_del(dpif, bond_id);
+    }
+
+    return error;
+}
+
+int
+dpif_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
+                    uint64_t *n_bytes)
+{
+    int error = -1;
+
+    if (dpif && dpif->dpif_class && dpif->dpif_class->bond_stats_get) {
+        error = dpif->dpif_class->bond_stats_get(dpif, bond_id, n_bytes);
+    }
+
+    return error;
+}
diff --git a/lib/dpif.h b/lib/dpif.h
index a88851e..6b3216f 100644
--- a/lib/dpif.h
+++ b/lib/dpif.h
@@ -908,6 +908,13 @@  char *dpif_get_dp_version(const struct dpif *);
 bool dpif_supports_tnl_push_pop(const struct dpif *);
 bool dpif_supports_explicit_drop_action(const struct dpif *);
 
+bool dpif_supports_lb_output_action(const struct dpif *);
+
+int dpif_bond_add(struct dpif *dpif, uint32_t bond_id, odp_port_t slave_map[]);
+int dpif_bond_del(struct dpif *dpif, uint32_t bond_id);
+int dpif_bond_stats_get(struct dpif *dpif, uint32_t bond_id,
+                        uint64_t *n_bytes);
+
 /* Log functions. */
 struct vlog_module;
 
diff --git a/lib/odp-execute.c b/lib/odp-execute.c
index 42d3335..6018e37 100644
--- a/lib/odp-execute.c
+++ b/lib/odp-execute.c
@@ -793,6 +793,7 @@  requires_datapath_assistance(const struct nlattr *a)
     switch (type) {
         /* These only make sense in the context of a datapath. */
     case OVS_ACTION_ATTR_OUTPUT:
+    case OVS_ACTION_ATTR_LB_OUTPUT:
     case OVS_ACTION_ATTR_TUNNEL_PUSH:
     case OVS_ACTION_ATTR_TUNNEL_POP:
     case OVS_ACTION_ATTR_USERSPACE:
@@ -1068,6 +1069,7 @@  odp_execute_actions(void *dp, struct dp_packet_batch *batch, bool steal,
             return;
         }
         case OVS_ACTION_ATTR_OUTPUT:
+        case OVS_ACTION_ATTR_LB_OUTPUT:
         case OVS_ACTION_ATTR_TUNNEL_PUSH:
         case OVS_ACTION_ATTR_TUNNEL_POP:
         case OVS_ACTION_ATTR_USERSPACE:
diff --git a/lib/odp-util.c b/lib/odp-util.c
index d6cb9e6..3472a8c 100644
--- a/lib/odp-util.c
+++ b/lib/odp-util.c
@@ -119,6 +119,7 @@  odp_action_len(uint16_t type)
 
     switch ((enum ovs_action_attr) type) {
     case OVS_ACTION_ATTR_OUTPUT: return sizeof(uint32_t);
+    case OVS_ACTION_ATTR_LB_OUTPUT: return sizeof(uint32_t);
     case OVS_ACTION_ATTR_TRUNC: return sizeof(struct ovs_action_trunc);
     case OVS_ACTION_ATTR_TUNNEL_PUSH: return ATTR_LEN_VARIABLE;
     case OVS_ACTION_ATTR_TUNNEL_POP: return sizeof(uint32_t);
@@ -1122,6 +1123,9 @@  format_odp_action(struct ds *ds, const struct nlattr *a,
     case OVS_ACTION_ATTR_OUTPUT:
         odp_portno_name_format(portno_names, nl_attr_get_odp_port(a), ds);
         break;
+    case OVS_ACTION_ATTR_LB_OUTPUT:
+        ds_put_format(ds, "lb_output(bond,%"PRIu32")", nl_attr_get_u32(a));
+        break;
     case OVS_ACTION_ATTR_TRUNC: {
         const struct ovs_action_trunc *trunc =
                        nl_attr_get_unspec(a, sizeof *trunc);
diff --git a/ofproto/bond.c b/ofproto/bond.c
index 405202f..d540635 100644
--- a/ofproto/bond.c
+++ b/ofproto/bond.c
@@ -54,10 +54,6 @@  static struct ovs_rwlock rwlock = OVS_RWLOCK_INITIALIZER;
 static struct hmap all_bonds__ = HMAP_INITIALIZER(&all_bonds__);
 static struct hmap *const all_bonds OVS_GUARDED_BY(rwlock) = &all_bonds__;
 
-/* Bit-mask for hashing a flow down to a bucket. */
-#define BOND_MASK 0xff
-#define BOND_BUCKETS (BOND_MASK + 1)
-
 /* Priority for internal rules created to handle recirculation */
 #define RECIRC_RULE_PRIORITY 20
 
@@ -126,6 +122,8 @@  struct bond {
     enum lacp_status lacp_status; /* Status of LACP negotiations. */
     bool bond_revalidate;       /* True if flows need revalidation. */
     uint32_t basis;             /* Basis for flow hash function. */
+    bool use_bond_cache;        /* Use bond cache to avoid recirculation.
+                                   Applicable only for Balance TCP mode. */
 
     /* SLB specific bonding info. */
     struct bond_entry *hash;     /* An array of BOND_BUCKETS elements. */
@@ -185,7 +183,7 @@  static struct bond_slave *choose_output_slave(const struct bond *,
                                               struct flow_wildcards *,
                                               uint16_t vlan)
     OVS_REQ_RDLOCK(rwlock);
-static void update_recirc_rules__(struct bond *bond);
+static void update_recirc_rules__(struct bond *bond, uint32_t bond_recirc_id);
 static bool bond_is_falling_back_to_ab(const struct bond *);
 
 /* Attempts to parse 's' as the name of a bond balancing mode.  If successful,
@@ -262,6 +260,7 @@  void
 bond_unref(struct bond *bond)
 {
     struct bond_slave *slave;
+    uint32_t bond_recirc_id = 0;
 
     if (!bond || ovs_refcount_unref_relaxed(&bond->ref_cnt) != 1) {
         return;
@@ -282,12 +281,13 @@  bond_unref(struct bond *bond)
 
     /* Free bond resources. Remove existing post recirc rules. */
     if (bond->recirc_id) {
+        bond_recirc_id = bond->recirc_id;
         recirc_free_id(bond->recirc_id);
         bond->recirc_id = 0;
     }
     free(bond->hash);
     bond->hash = NULL;
-    update_recirc_rules__(bond);
+    update_recirc_rules__(bond, bond_recirc_id);
 
     hmap_destroy(&bond->pr_rule_ops);
     free(bond->name);
@@ -328,13 +328,14 @@  add_pr_rule(struct bond *bond, const struct match *match,
  * lock annotation. Currently, only 'bond_unref()' calls
  * this function directly.  */
 static void
-update_recirc_rules__(struct bond *bond)
+update_recirc_rules__(struct bond *bond, uint32_t bond_recirc_id)
 {
     struct match match;
     struct bond_pr_rule_op *pr_op, *next_op;
     uint64_t ofpacts_stub[128 / 8];
     struct ofpbuf ofpacts;
     int i;
+    ofp_port_t slave_map[BOND_MASK];
 
     ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
 
@@ -353,8 +354,18 @@  update_recirc_rules__(struct bond *bond)
 
                 add_pr_rule(bond, &match, slave->ofp_port,
                             &bond->hash[i].pr_rule);
+                slave_map[i] = slave->ofp_port;
+            } else {
+                slave_map[i] = OFP_PORT_C(-1);
             }
         }
+        if (bond->ofproto->backer->rt_support.lb_output_action) {
+            ofproto_dpif_bundle_add(bond->ofproto, bond->recirc_id, slave_map);
+        }
+    } else {
+        if (bond->ofproto->backer->rt_support.lb_output_action) {
+            ofproto_dpif_bundle_del(bond->ofproto, bond_recirc_id);
+        }
     }
 
     HMAP_FOR_EACH_SAFE(pr_op, next_op, hmap_node, &bond->pr_rule_ops) {
@@ -404,7 +415,7 @@  static void
 update_recirc_rules(struct bond *bond)
     OVS_REQ_RDLOCK(rwlock)
 {
-    update_recirc_rules__(bond);
+    update_recirc_rules__(bond, bond->recirc_id);
 }
 
 /* Updates 'bond''s overall configuration to 's'.
@@ -467,6 +478,10 @@  bond_reconfigure(struct bond *bond, const struct bond_settings *s)
         recirc_free_id(bond->recirc_id);
         bond->recirc_id = 0;
     }
+    if (bond->use_bond_cache != s->use_bond_cache) {
+        bond->use_bond_cache = s->use_bond_cache;
+        revalidate = true;
+    }
 
     if (bond->balance == BM_AB || !bond->hash || revalidate) {
         bond_entry_reset(bond);
@@ -944,6 +959,15 @@  bond_recirculation_account(struct bond *bond)
     OVS_REQ_WRLOCK(rwlock)
 {
     int i;
+    uint64_t n_bytes[BOND_BUCKETS] = {0};
+
+    if (bond->hash && bond->recirc_id) {
+        /* Retrieve bond stats from datapath. */
+        if (bond->ofproto->backer->rt_support.lb_output_action) {
+            dpif_bond_stats_get(bond->ofproto->backer->dpif,
+                                bond->recirc_id, n_bytes);
+       }
+    }
 
     for (i=0; i<=BOND_MASK; i++) {
         struct bond_entry *entry = &bond->hash[i];
@@ -953,8 +977,12 @@  bond_recirculation_account(struct bond *bond)
             struct pkt_stats stats;
             long long int used OVS_UNUSED;
 
-            rule->ofproto->ofproto_class->rule_get_stats(
-                rule, &stats, &used);
+            if (!bond->ofproto->backer->rt_support.lb_output_action) {
+                rule->ofproto->ofproto_class->rule_get_stats(
+                    rule, &stats, &used);
+            } else {
+                stats.n_bytes = n_bytes[i];
+            }
             bond_entry_account(entry, stats.n_bytes);
         }
     }
@@ -1365,6 +1393,8 @@  bond_print_details(struct ds *ds, const struct bond *bond)
                   may_recirc ? "yes" : "no", may_recirc ? recirc_id: -1);
 
     ds_put_format(ds, "bond-hash-basis: %"PRIu32"\n", bond->basis);
+    ds_put_format(ds, "lb-output-action: %s\n",
+                  bond->use_bond_cache ? "enabled" : "disabled");
 
     ds_put_format(ds, "updelay: %d ms\n", bond->updelay);
     ds_put_format(ds, "downdelay: %d ms\n", bond->downdelay);
@@ -1942,3 +1972,9 @@  bond_get_changed_active_slave(const char *name, struct eth_addr *mac,
 
     return false;
 }
+
+bool
+bond_get_cache_mode(const struct bond *bond)
+{
+    return bond->use_bond_cache;
+}
diff --git a/ofproto/bond.h b/ofproto/bond.h
index e7c3d9b..88a4de1 100644
--- a/ofproto/bond.h
+++ b/ofproto/bond.h
@@ -22,6 +22,10 @@ 
 #include "ofproto-provider.h"
 #include "packets.h"
 
+/* Bit-mask for hashing a flow down to a bucket. */
+#define BOND_MASK 0xff
+#define BOND_BUCKETS (BOND_MASK + 1)
+
 struct flow;
 struct netdev;
 struct ofpbuf;
@@ -58,6 +62,8 @@  struct bond_settings {
                                 /* The MAC address of the interface
                                    that was active during the last
                                    ovs run. */
+    bool use_bond_cache;        /* Use bond cache. Only applicable for
+                                   bond mode BALANCE TCP. */
 };
 
 /* Program startup. */
@@ -122,4 +128,7 @@  void bond_rebalance(struct bond *);
 */
 void bond_update_post_recirc_rules(struct bond *, uint32_t *recirc_id,
                                    uint32_t *hash_basis);
+
+bool bond_get_cache_mode(const struct bond *);
+
 #endif /* bond.h */
diff --git a/ofproto/ofproto-dpif-ipfix.c b/ofproto/ofproto-dpif-ipfix.c
index b413768..5030978 100644
--- a/ofproto/ofproto-dpif-ipfix.c
+++ b/ofproto/ofproto-dpif-ipfix.c
@@ -3017,6 +3017,7 @@  dpif_ipfix_read_actions(const struct flow *flow,
         case OVS_ACTION_ATTR_CHECK_PKT_LEN:
         case OVS_ACTION_ATTR_UNSPEC:
         case OVS_ACTION_ATTR_DROP:
+        case OVS_ACTION_ATTR_LB_OUTPUT:
         case __OVS_ACTION_ATTR_MAX:
         default:
             break;
diff --git a/ofproto/ofproto-dpif-sflow.c b/ofproto/ofproto-dpif-sflow.c
index f9ea47a..21dafa6 100644
--- a/ofproto/ofproto-dpif-sflow.c
+++ b/ofproto/ofproto-dpif-sflow.c
@@ -1177,6 +1177,7 @@  dpif_sflow_read_actions(const struct flow *flow,
         case OVS_ACTION_ATTR_CT:
     case OVS_ACTION_ATTR_CT_CLEAR:
         case OVS_ACTION_ATTR_METER:
+        case OVS_ACTION_ATTR_LB_OUTPUT:
             break;
 
         case OVS_ACTION_ATTR_SET_MASKED:
diff --git a/ofproto/ofproto-dpif-xlate.c b/ofproto/ofproto-dpif-xlate.c
index 4407f9c..0446387 100644
--- a/ofproto/ofproto-dpif-xlate.c
+++ b/ofproto/ofproto-dpif-xlate.c
@@ -409,6 +409,8 @@  struct xlate_ctx {
     struct ofpbuf action_set;   /* Action set. */
 
     enum xlate_error error;     /* Translation failed. */
+
+    bool tnl_push_no_recirc;    /* Tunnel push recirculation status */
 };
 
 /* Structure to track VLAN manipulation */
@@ -3727,12 +3729,16 @@  native_tunnel_output(struct xlate_ctx *ctx, const struct xport *xport,
         ctx->xin->allow_side_effects = backup_side_effects;
         ctx->xin->packet = backup_packet;
         ctx->wc = backup_wc;
+
+        ctx->tnl_push_no_recirc = true;
     } else {
         /* In order to maintain accurate stats, use recirc for
          * natvie tunneling.  */
         nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, 0);
         nl_msg_end_nested(ctx->odp_actions, clone_ofs);
-    }
+
+        ctx->tnl_push_no_recirc = false;
+   }
 
     /* Restore the flows after the translation. */
     memcpy(&ctx->xin->flow, &old_flow, sizeof ctx->xin->flow);
@@ -4183,7 +4189,6 @@  compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
         xlate_commit_actions(ctx);
 
         if (xr) {
-            /* Recirculate the packet. */
             struct ovs_action_hash *act_hash;
 
             /* Hash action. */
@@ -4192,15 +4197,30 @@  compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
                 /* Algorithm supported by all datapaths. */
                 hash_alg = OVS_HASH_ALG_L4;
             }
-            act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
+
+            if ((xr->hash_alg == OVS_HASH_ALG_L4) &&
+                (ctx->xbridge->support.lb_output_action) &&
+                !ctx->tnl_push_no_recirc &&
+                bond_get_cache_mode(xport->xbundle->bond)) {
+                /*
+                 * If bond mode is balance-tcp and optimize balance tcp
+                 * is enabled then use the hash directly for slave selection
+                 * and avoid recirculation.
+                 *
+                 * Currently support for netdev datapath only.
+                 */
+                nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_LB_OUTPUT,
+                               xr->recirc_id);
+            } else {
+                act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
                                                 OVS_ACTION_ATTR_HASH,
                                                 sizeof *act_hash);
-            act_hash->hash_alg = hash_alg;
-            act_hash->hash_basis = xr->hash_basis;
-
-            /* Recirc action. */
-            nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
-                           xr->recirc_id);
+                act_hash->hash_alg = hash_alg;
+                act_hash->hash_basis = xr->hash_basis;
+                /* Recirculate the packet. */
+                nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
+                               xr->recirc_id);
+            }
         } else if (is_native_tunnel) {
             /* Output to native tunnel port. */
             native_tunnel_output(ctx, xport, flow, odp_port, truncate);
@@ -7261,7 +7281,8 @@  count_output_actions(const struct ofpbuf *odp_actions)
     int n = 0;
 
     NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
-        if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
+        if ((a->nla_type == OVS_ACTION_ATTR_OUTPUT) ||
+            (a->nla_type == OVS_ACTION_ATTR_LB_OUTPUT)) {
             n++;
         }
     }
@@ -7458,6 +7479,7 @@  xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
 
         .action_set_has_group = false,
         .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
+        .tnl_push_no_recirc = false,
     };
 
     /* 'base_flow' reflects the packet as it came in, but we need it to reflect
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index f782f86..3c5438b 100644
--- a/ofproto/ofproto-dpif.c
+++ b/ofproto/ofproto-dpif.c
@@ -1580,6 +1580,8 @@  check_support(struct dpif_backer *backer)
     backer->rt_support.ct_timeout = check_ct_timeout_policy(backer);
     backer->rt_support.explicit_drop_action =
         dpif_supports_explicit_drop_action(backer->dpif);
+    backer->rt_support.lb_output_action=
+        dpif_supports_lb_output_action(backer->dpif);
 
     /* Flow fields. */
     backer->rt_support.odp.ct_state = check_ct_state(backer);
@@ -3433,6 +3435,27 @@  bundle_remove(struct ofport *port_)
     }
 }
 
+int
+ofproto_dpif_bundle_add(struct ofproto_dpif *ofproto,
+                        uint32_t bond_id, const ofp_port_t slave_map[])
+{
+    /* Convert ofp_port to odp_port */
+    odp_port_t odp_map[BOND_BUCKETS];
+    for (int bucket = 0; bucket < BOND_BUCKETS; bucket++) {
+        odp_map[bucket] = (slave_map[bucket] == OFP_PORT_C(-1)
+                           ? ODP_PORT_C(-1)
+                           : ofp_port_to_odp_port(ofproto, slave_map[bucket]));
+    }
+
+    return dpif_bond_add(ofproto->backer->dpif, bond_id, odp_map);
+}
+
+int
+ofproto_dpif_bundle_del(struct ofproto_dpif *ofproto, uint32_t bond_id)
+{
+    return dpif_bond_del(ofproto->backer->dpif, bond_id);
+}
+
 static void
 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
 {
@@ -5563,6 +5586,7 @@  get_datapath_cap(const char *datapath_type, struct smap *cap)
     smap_add(cap, "ct_timeout", s.ct_timeout ? "true" : "false");
     smap_add(cap, "explicit_drop_action",
              s.explicit_drop_action ? "true" :"false");
+    smap_add(cap, "lb_output_action", s.lb_output_action ? "true" : "false");
 }
 
 /* Gets timeout policy name in 'backer' based on 'zone', 'dl_type' and
diff --git a/ofproto/ofproto-dpif.h b/ofproto/ofproto-dpif.h
index c9d5df3..384a269 100644
--- a/ofproto/ofproto-dpif.h
+++ b/ofproto/ofproto-dpif.h
@@ -202,8 +202,10 @@  struct group_dpif *group_dpif_lookup(struct ofproto_dpif *,
     DPIF_SUPPORT_FIELD(bool, ct_timeout, "Conntrack timeout policy")        \
                                                                             \
     /* True if the datapath supports explicit drop action. */               \
-    DPIF_SUPPORT_FIELD(bool, explicit_drop_action, "Explicit Drop action")
-
+    DPIF_SUPPORT_FIELD(bool, explicit_drop_action, "Explicit Drop action")  \
+                                                                            \
+    /* True if the datapath supports balance_tcp optimization */            \
+    DPIF_SUPPORT_FIELD(bool, lb_output_action, "Optimized Balance TCP mode")
 
 /* Stores the various features which the corresponding backer supports. */
 struct dpif_backer_support {
@@ -379,6 +381,9 @@  int ofproto_dpif_add_internal_flow(struct ofproto_dpif *,
                                    struct rule **rulep);
 int ofproto_dpif_delete_internal_flow(struct ofproto_dpif *, struct match *,
                                       int priority);
+int ofproto_dpif_bundle_add(struct ofproto_dpif *, uint32_t bond_id,
+                            const ofp_port_t slave_map[]);
+int ofproto_dpif_bundle_del(struct ofproto_dpif *, uint32_t bond_id);
 
 bool ovs_native_tunneling_is_on(struct ofproto_dpif *);
 
diff --git a/tests/lacp.at b/tests/lacp.at
index 7b460d7..c881a0b 100644
--- a/tests/lacp.at
+++ b/tests/lacp.at
@@ -121,6 +121,7 @@  AT_CHECK([ovs-appctl bond/show], [0], [dnl
 bond_mode: active-backup
 bond may use recirculation: no, Recirc-ID : -1
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -286,6 +287,7 @@  slave: p3: current attached
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -301,6 +303,7 @@  slave p1: enabled
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -423,6 +426,7 @@  slave: p3: current attached
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -440,6 +444,7 @@  slave p1: enabled
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -555,6 +560,7 @@  slave: p3: current attached
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -572,6 +578,7 @@  slave p1: enabled
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -692,6 +699,7 @@  slave: p3: current attached
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
@@ -709,6 +717,7 @@  slave p1: enabled
 bond_mode: balance-tcp
 bond may use recirculation: yes, <del>
 bond-hash-basis: 0
+lb-output-action: disabled
 updelay: 0 ms
 downdelay: 0 ms
 lacp_status: negotiated
diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
index 86c7b10..d4f34ad 100644
--- a/vswitchd/bridge.c
+++ b/vswitchd/bridge.c
@@ -4589,6 +4589,12 @@  port_configure_bond(struct port *port, struct bond_settings *s)
         /* OVSDB did not store the last active interface */
         s->active_slave_mac = eth_addr_zero;
     }
+    /* Default set bond cache use to false. */
+    s->use_bond_cache = false;
+    if (s->balance == BM_TCP) {
+        s->use_bond_cache = smap_get_bool(&port->cfg->other_config,
+                                        "lb-output-action", false);
+    }
 }
 
 /* Returns true if 'port' is synthetic, that is, if we constructed it locally
diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
index 0ec726c..bdc7ef4 100644
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -1962,6 +1962,16 @@ 
         <code>active-backup</code>.
       </column>
 
+      <column name="other_config" key="lb-output-action"
+              type='{"type": "boolean"}'>
+        Enable/disable usage of optimized lb-output-action for
+        balancing flows among output slaves in load balanced bonds in
+        <code>balance-tcp</code>. When enabled, it uses optimized path for
+        balance-tcp mode by using rss hash and avoids recirculation.
+        It affects only new flows, i.e, existing flows remain unchanged.
+        This knob does not affect other balancing modes.
+      </column>
+
       <group title="Link Failure Detection">
         <p>
           An important part of link bonding is detecting that links are down so
@@ -5722,6 +5732,19 @@  ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
         higher performance for MPLS and active-active load balancing
         bonding modes.
       </column>
+      <column name="capabilities" key="lb_output_action"
+              type='{"type": "boolean"}'>
+        If this is true, then the datapath supports optimized balance-tcp
+        bond mode. This capability replaces existing <code>hash</code> and
+        <code>recirc</code> actions with new action <code>lb_output</code>
+        and avoids recirculation of packet in datapath. It is supported
+        only for balance-tcp bond mode in netdev datapath. The new action
+        gives higer performance but also requires pmd reloads whenever
+        traffic load needs to be rebalanced among bonded links. By default
+        this new action is disabled, however it can be enabled by setting
+        <ref column="other-config" key="lb-output-action"/> in
+        <ref table="Port"/> table.
+      </column>
       <group title="Connection-Tracking Capabilities">
         <p>
           These capabilities are granular because Open vSwitch and its