[v2] net/i40e: fix request queue fail in VF

56748 diff mbox series

From: Zhu Tao <taox.zhu@intel.com>

When the VF configuration is larger than the number of queues reserved
by PF, VF sends the request queue command through admin queue. When PF
received this command, it may reset the VF and send a notification
before resetting. If this notification is read by the timed task alarm,
Task request queue will lost notification. This patch Mark vf_reset,
pend_msg flag just as task request queue has received notification in
task alarm.

Fixes: 864a800d70 ("net/i40e: remove VF interrupt handler")
Fixes: ee653bd800 ("net/i40e: determine number of queues per VF at run time")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/i40e/i40e_ethdev_vf.c | 4 ++++
 1 file changed, 4 insertions(+)

> -----Original Message-----
> From: Zhu, TaoX
> Sent: Friday, July 19, 2019 11:18 AM
> To: Xing, Beilei <beilei.xing@intel.com>; Zhang, Qi Z <qi.z.zhang@intel.com>
> Cc: dev@dpdk.org; Zhu, TaoX <taox.zhu@intel.com>; stable@dpdk.org
> Subject: [PATCH v2] net/i40e: fix request queue fail in VF
> 
> From: Zhu Tao <taox.zhu@intel.com>
> 
> When the VF configuration is larger than the number of queues reserved by PF,
> VF sends the request queue command through admin queue. When PF
> received this command, it may reset the VF and send a notification before
> resetting. If this notification is read by the timed task alarm, Task request
> queue will lost notification. This patch Mark vf_reset, pend_msg flag just as
> task request queue has received notification in task alarm.
> 
> Fixes: 864a800d70 ("net/i40e: remove VF interrupt handler")
> Fixes: ee653bd800 ("net/i40e: determine number of queues per VF at run
> time")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zhu Tao <taox.zhu@intel.com>
> ---
>  drivers/net/i40e/i40e_ethdev_vf.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/net/i40e/i40e_ethdev_vf.c
> b/drivers/net/i40e/i40e_ethdev_vf.c
> index 5be32b069..3b9740c08 100644
> --- a/drivers/net/i40e/i40e_ethdev_vf.c
> +++ b/drivers/net/i40e/i40e_ethdev_vf.c
> @@ -1332,6 +1332,10 @@ i40evf_handle_pf_event(struct rte_eth_dev *dev,
> uint8_t *msg,
>  		PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_RESET_IMPENDING
> event");
>  		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
>  					      NULL);
> +		if (!vf->vf_reset) {
> +			vf->vf_reset = true;
> +			vf->pend_msg|= PFMSG_RESET_IMPENDING;

You current solution rely on we can always can get a dummy event after the VIRTCHNL_EVENT_RESET_IMPENDING event we needed, but I'm not sure if it is guaranteed

If the issue is due to the race condition between i40evf_execute_vf_cmd and i40evf_dev_alarm_handler
Why not just disable the adminq interrupt before i40evf_request_queues and enable it back after it?

> +		}
>  		break;
>  	case VIRTCHNL_EVENT_LINK_CHANGE:
>  		PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_LINK_CHANGE event");
> --
> 2.17.1