patch: allow bonding slaves from different numa nodes

Note the patch may have a negative influnce on performance.
It's not a good practice to bonding slaves across numa nodes.

Signed-off-by: ywc689 <[email protected]>

patch/dpdk-stable-20.11.1/0001-kni-use-netlink-event-for-multicast-driver-part.patch

@@ -1,7 +1,7 @@
-From 8e66b7600c7d8d0e59dd85565986436c0e03a2c8 Mon Sep 17 00:00:00 2001
+From 09760dd24ceaa9a6ce2435cdf9105e1da5d95616 Mon Sep 17 00:00:00 2001
 From: huangyichen <[email protected]>
 Date: Thu, 1 Jul 2021 21:21:16 +0800
-Subject: [PATCH 1/5] kni: use netlink event for multicast (driver part)
+Subject: [PATCH 1/6] kni: use netlink event for multicast (driver part)
 
 Kni driver sends netlink event every time hw-multicast list updated by
 kernel, the user kni app should capture the event and update multicast

patch/dpdk-stable-20.11.1/0002-pdump-change-dpdk-pdump-tool-for-dpvs.patch

@@ -1,7 +1,7 @@
-From 0ef2e126c8ca9b4a246f680a2a5110b734e9782b Mon Sep 17 00:00:00 2001
+From 60e1eb0a50fc10ca69bcdc4148fdec35576eeedc Mon Sep 17 00:00:00 2001
 From: huangyichen <[email protected]>
 Date: Thu, 1 Jul 2021 21:23:50 +0800
-Subject: [PATCH 2/5] pdump: change dpdk-pdump tool for dpvs
+Subject: [PATCH 2/6] pdump: change dpdk-pdump tool for dpvs
 
 ---
  app/pdump/main.c             | 167 ++++++++++++++++++++++++++++++++++++++++---

patch/dpdk-stable-20.11.1/0003-for-debug-only-enable-dpdk-eal-memory-debug.patch

@@ -1,7 +1,7 @@
-From 446e9d2ae65c25ca382323bdd7cdec765f357886 Mon Sep 17 00:00:00 2001
+From c9763a0b83bdf0fbf9a5a0ca8c96d999cca62196 Mon Sep 17 00:00:00 2001
 From: huangyichen <[email protected]>
 Date: Thu, 1 Jul 2021 21:24:47 +0800
-Subject: [PATCH 3/5] [for debug only] enable dpdk eal memory debug
+Subject: [PATCH 3/6] [for debug only] enable dpdk eal memory debug
 
 The patch is used for memory debug. To use the patch, configure meson with option
 -Dc_args="-DRTE_MALLOC_DEBUG" when building dpdk. For example,

patch/dpdk-stable-20.11.1/0004-ixgbe_flow-patch-ixgbe-fdir-rte_flow-for-dpvs.patch

@@ -1,7 +1,7 @@
-From fc81d45d6bf23bd6d9b0e7eb9fdc7fe41b559e65 Mon Sep 17 00:00:00 2001
+From 54aad262b48f5904d7f1688b56526908151f5574 Mon Sep 17 00:00:00 2001
 From: huangyichen <[email protected]>
 Date: Fri, 2 Jul 2021 11:55:47 +0800
-Subject: [PATCH 4/5] ixgbe_flow: patch ixgbe fdir rte_flow for dpvs
+Subject: [PATCH 4/6] ixgbe_flow: patch ixgbe fdir rte_flow for dpvs
 
 1. Ignore fdir flow rule priority attribute.
 2. Use different fdir soft-id for flow rules configured for the same queue.

patch/dpdk-stable-20.11.1/0005-Fix-bonding-mode-4-problem-caused-by-LACP-failure.patch

@@ -1,7 +1,7 @@
-From ad725a88c0acbd795646cf1d6280c5de7a012044 Mon Sep 17 00:00:00 2001
+From 0d0ba57681393cd9c11a0ac069c6f676b9f34707 Mon Sep 17 00:00:00 2001
 From: huangyichen <[email protected]>
 Date: Fri, 2 Jul 2021 11:57:03 +0800
-Subject: [PATCH 5/5] Fix bonding mode 4 problem caused by LACP failure.
+Subject: [PATCH 5/6] Fix bonding mode 4 problem caused by LACP failure.
 
 The problem is disscussed in Issue #725 of iqiyi/dpvs in detail.
 https://github.com/iqiyi/dpvs/issues/725

patch/dpdk-stable-20.11.1/0006-bonding-allow-slaves-from-different-numa-nodes.patch

@@ -0,0 +1,52 @@
+From e6d97675f818b6225209a3cada1e53a0756daedb Mon Sep 17 00:00:00 2001
+From: wencyu <[email protected]>
+Date: Mon, 2 Aug 2021 13:52:24 +0800
+Subject: [PATCH 6/6] bonding: allow slaves from different numa nodes
+
+Note the patch may have a negative influnce on performance.
+It's not a good practice to bonding slaves across numa nodes.
+
+Signed-off-by: wencyu <[email protected]>
+---
+ drivers/net/bonding/rte_eth_bond_pmd.c | 18 ++++++++++++++++--
+ 1 file changed, 16 insertions(+), 2 deletions(-)
+
+diff --git a/drivers/net/bonding/rte_eth_bond_pmd.c b/drivers/net/bonding/rte_eth_bond_pmd.c
+index 057b1ad..53f8ba3 100644
+--- a/drivers/net/bonding/rte_eth_bond_pmd.c
++++ b/drivers/net/bonding/rte_eth_bond_pmd.c
+@@ -1762,7 +1762,14 @@ slave_configure(struct rte_eth_dev *bonded_eth_dev,
+
+ 		errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
+ 				bd_rx_q->nb_rx_desc,
+-				rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
++				// In spite of performance problem, bonding slaves had better to support
++				// slaves from different numa nodes. Considering that numa node on which
++				// the resources of bonding port is allocated from is specified by
++				// rte_eth_bond_create() at bonding creation, the slave's queue_setup
++				// would fail if specified with the slave's numa node id that is different
++				// from the one of the bonding port. See rte_eth_dma_zone_reserve() for
++				// details.
++				SOCKET_ID_ANY,
+ 				&(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
+ 		if (errval != 0) {
+ 			RTE_BOND_LOG(ERR,
+@@ -1778,7 +1785,14 @@ slave_configure(struct rte_eth_dev *bonded_eth_dev,
+
+ 		errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
+ 				bd_tx_q->nb_tx_desc,
+-				rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
++				// In spite of performance problem, bonding slaves had better to support
++				// slaves from different numa nodes. Considering that numa node on which
++				// the resources of bonding port is allocated from is specified by
++				// rte_eth_bond_create() at bonding creation, the slave's queue_setup
++				// would fail if specified with the slave's numa node id that is different
++				// from the one of the bonding port. See rte_eth_dma_zone_reserve() for
++				// details.
++				SOCKET_ID_ANY,
+ 				&bd_tx_q->tx_conf);
+ 		if (errval != 0) {
+ 			RTE_BOND_LOG(ERR,
+-- 
+1.8.3.1
+