Thread (9 messages) 9 messages, 3 authors, 2017-02-09

Re: [PATCH v2] net/i40e: set no drop for traffic class

From: Wu, Jingjing <hidden>
Date: 2017-02-07 15:26:00

-----Original Message-----
From: Sexton, Rory
Sent: Thursday, January 19, 2017 6:38 PM
To: Wu, Jingjing <redacted>
Cc: dev@dpdk.org; Marjanovic, Nemanja <redacted>
Subject: RE: [PATCH v2] net/i40e: set no drop for traffic class

Perhaps the best solution is as suggested to set rte_eth_conf.dcb_capability_en
= ETH_DCB_PFC_SUPPORT rte_eth_conf.rxmode.mq_mode =
ETH_MQ_RX_DCB_FLAG and set rte_eth_dcb_rx_conf.nb_tcs to the number of
tc's to apply Using this port configuration will give the same behavior of the
patch and it removes the need for an API change.

Rory
Yes, That's what I thought when the v1 patch. So do we still need this patch now?
-----Original Message-----
From: Wu, Jingjing
Sent: Tuesday, January 17, 2017 3:09 PM
To: Sexton, Rory <redacted>
Cc: dev@dpdk.org; Marjanovic, Nemanja <redacted>
Subject: RE: [PATCH v2] net/i40e: set no drop for traffic class


quoted
-----Original Message-----
From: Sexton, Rory
Sent: Monday, January 16, 2017 11:52 PM
To: Wu, Jingjing <redacted>
Cc: dev@dpdk.org; Sexton, Rory <redacted>; Marjanovic,
Nemanja [off-list ref]
Subject: [PATCH v2] net/i40e: set no drop for traffic class

From: Rory Sexton <redacted>

The default traffic class in i40e is set to drop versus on ixgbe it
isset to no drop. This means when packets build up in the RX SRAM on
the NIC, they are dropped, and they do this when the SW descriptor rings fill up.

This patch changes this behaviour and our testing shows there are no
drops as a result.

Signed-off-by: Rory Sexton <redacted>
Signed-off-by: Nemanja Marjanovic <redacted>
---
v2:
* Changed to use existing api to set priority register directly.

 drivers/net/i40e/i40e_ethdev.c | 3 +++
 1 file changed, 3 insertions(+)
diff --git a/drivers/net/i40e/i40e_ethdev.c
b/drivers/net/i40e/i40e_ethdev.c index 67778ba..97339b5 100644
--- a/drivers/net/i40e/i40e_ethdev.c
+++ b/drivers/net/i40e/i40e_ethdev.c
@@ -2985,8 +2985,11 @@ static int
 i40e_priority_flow_ctrl_set(__rte_unused struct rte_eth_dev *dev,
 			    __rte_unused struct rte_eth_pfc_conf *pfc_conf) {
+	struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data-
quoted
dev_private);
+
 	PMD_INIT_FUNC_TRACE();

+	I40E_WRITE_REG(hw, I40E_PRTDCB_TC2PFC, 0xff);
PRTDCB_TC2PFC  is the Bitmap who controls the use of Priority Flow Control
(PFC) per each TC. Bit n set to 1b indicates TC n uses PFC in Rx and Tx. The TC is
referred as a no-drop TC.

And if look the rte_eth_pfc_conf, there is a field called priority, which would
map to a TC.
Currently, the TC and priority is 1:1 map when dcb is enabled.
So how about change it like:
Check dcb info, and map the priority to tc, then val = 0x1 << tc;
I40E_WRITE_REG(hw, I40E_PRTDCB_TC2PFC, val);

Thanks
Jingjing


quoted
 	return -ENOSYS;
 }

--
2.4.3
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help