[dpdk-dev] [PATCH v4] net/ixgbe: fix l3fwd start failed on VF

Wu, Yanglong yanglong.wu at intel.com
Fri Jan 12 03:38:22 CET 2018


Hi
After confirm with Daiwei and according to the data sheet, RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool is the Max number of queue can be  used in VF, that would be assigned as IXGBE_MAX_RX_QUEUE_NUM / RTE_ETH_DEV_SRIOV(dev).active, so that assigning nb_q_per_pool  as 1 when PF is in ETH_MQ_RX_NONE, which will make VF can just use only 1 queue, is not right. So this patch is right.

Yanglong

-----Original Message-----
From: Wu, Yanglong 
Sent: Thursday, January 11, 2018 11:29 PM
To: Wu, Yanglong <yanglong.wu at intel.com>; Dai, Wei <wei.dai at intel.com>; dev at dpdk.org
Subject: RE: [dpdk-dev] [PATCH v4] net/ixgbe: fix l3fwd start failed on VF

Hi daiwei:

After I double checked with this bug, the problem can be described as follow:
When we start PF using testpmd like this: testpmd -c 0x3 -n 4 --socket-mem=1024,1024 --file-prefix=pf -w 0000:05:00.0 -w 0000:05:00.1 -- -i. The number of queue signed to each port would be 1 and the  rxmode.mq_mode of ixgbe eth_dev  would be set as  ETH_MQ_RX_NONE. And this config info is make sense, so until now there is no problem, right?
And according to the code:
		case ETH_MQ_RX_NONE:
>  			/* if nothing mq mode configure, use default scheme */
>  			dev->data->dev_conf.rxmode.mq_mode = ETH_MQ_RX_VMDQ_ONLY;
> -			if (RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool > 1)
> -				RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
>  			break;
The nb_q_per_pool of (RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool  would be set as 1, that means for VF the number of max queue can be used would be only 1. 

And next we start VF using l3fwd:
l3fwd/build/l3fwd -c 0xc0000 -n 4 -w 0000:05:10.0 -w 0000:05:10.1 -- -p 0x3 --config '(0,0,18),(1,0,19)' --parse-ptype The number of rx_q equal to 1 and there is no problem. But, in l3fwd, the number of tx_q would equal to the number of core used in l3fwd, so here number of tx_q equal to 2 but as described above, the max number of queue can be used in VF is just only 1. So there is problem and l3fwd will check failed.  


So there is two ways to fix this bug, the first is using this patch, don't assign RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool as 1 when mq_mode = ETH_MQ_RX_VMDQ_ONLY.
And the second is that the number of tx_q would not assigned as the number of cores used in l3fwd.

Which one do you think is make sense? 




Yanglong


-----Original Message-----
From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Wu, Yanglong
Sent: Thursday, January 11, 2018 6:30 PM
To: Dai, Wei <wei.dai at intel.com>; dev at dpdk.org
Subject: Re: [dpdk-dev] [PATCH v4] net/ixgbe: fix l3fwd start failed on VF

Yes, this problem is caused by PF passing a wrong number of queue to VF. So basically, I removed  model check of "case ETH_MQ_RX_NONE". It seem right after I do that but I'am also not very sure the changing is really right. So thank to your comments, I will redebug this problem.

-----Original Message-----
From: Dai, Wei
Sent: Thursday, January 11, 2018 5:48 PM
To: Wu, Yanglong <yanglong.wu at intel.com>; dev at dpdk.org; Yang, Zhiyong <zhiyong.yang at intel.com>
Cc: Wu, Yanglong <yanglong.wu at intel.com>
Subject: RE: [dpdk-dev] [PATCH v4] net/ixgbe: fix l3fwd start failed on VF

According to 82599 datasheet, VF working mode is controlled by its host PF.
If VF is working in RSS or DCB mode, it can use more than 1 queues.
That also means its host PF should enable RSS or DCB.
The PF Rx mode can be ETH_MQ_RX_VMDQ_RSS or ETH_MQ_RX_VMDQ_DCB.
So I think that DPDK app with PF cause this issue, not ixgbe PMD itself.

By the way, you can reply your patch mail to provide more info about how You reproduce this issue to easy others' reviewing and testing your patch.

Thanks

> -----Original Message-----
> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Yanglong Wu
> Sent: Tuesday, January 9, 2018 2:32 PM
> To: dev at dpdk.org; Yang, Zhiyong <zhiyong.yang at intel.com>
> Cc: Wu, Yanglong <yanglong.wu at intel.com>
> Subject: [dpdk-dev] [PATCH v4] net/ixgbe: fix l3fwd start failed on VF
> 
> VF can't run in multi queue mode, if nb_q_per_pool was set as 1.
> Nb_q_per_pool is passed through to max_rx_q and max_tx_q in VF.
> So if nb_q_per_pool is equal to 1, max_rx_q and max_tx_q can't be more 
> than 1 and VF multi queue mode will fail.
> 
> Fixes: 27b609cbd1c6 (ethdev: move the multi-queue mode check)
> 
> Signed-off-by: Yanglong Wu <yanglong.wu at intel.com>
> ---
> v1~v3:
> coding style issue fix
> ---
> v4:
> Rework according to comments
> ---
>  drivers/net/ixgbe/ixgbe_ethdev.c | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c
> b/drivers/net/ixgbe/ixgbe_ethdev.c
> index baaeee5d9..6af4f8b2b 100644
> --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> @@ -2240,8 +2240,6 @@ ixgbe_check_mq_mode(struct rte_eth_dev *dev)
>  		case ETH_MQ_RX_NONE:
>  			/* if nothing mq mode configure, use default scheme */
>  			dev->data->dev_conf.rxmode.mq_mode = ETH_MQ_RX_VMDQ_ONLY;
> -			if (RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool > 1)
> -				RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
>  			break;
>  		default: /* ETH_MQ_RX_DCB, ETH_MQ_RX_DCB_RSS or ETH_MQ_TX_DCB*/
>  			/* SRIOV only works in VMDq enable mode */
> --
> 2.11.0



More information about the dev mailing list