> -----Original Message-----
> From: Yu, DapengX <dapengx...@intel.com>
> Sent: Tuesday, October 26, 2021 4:44 PM
> To: Yang, Qiming <qiming.y...@intel.com>; Zhang, Qi Z
> <qi.z.zh...@intel.com>
> Cc: dev@dpdk.org; Yu, DapengX <dapengx...@intel.com>; sta...@dpdk.org
> Subject: [PATCH v4] net/ice: simplify the use of DCF device reset
>
> From: Dapeng Yu <dapengx...@intel.com>
>
> After DCF is reset by PF, the DCF device un-initialization cannot function
> normally since the resource is already invalidated. So reset DCF twice is
> necessary, the first simplified reset re-initializes the AdminQ of DCF, only
> then
> second reset can clean the filters successfully.
>
> This patch detects the reset flag, which is set by PF on DCF reset, if the
> flag is
> true, do DCF reset twice automatically.
>
> Fixes: 1a86f4dbdf42 ("net/ice: support DCF device reset")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Dapeng Yu <dapengx...@intel.com>
> ---
> V2:
> * Ignore the returned error of dev_uninit when DCF is reset by PF
> V3:
> * Add a reset function to re-initialize AdminQ resource
> * Add a function to check the reset flag
> V4:
> * Remove redundant reset flag setting
> ---
> drivers/net/ice/ice_dcf.c | 2 ++
> drivers/net/ice/ice_dcf_ethdev.c | 33 +++++++++++++++++++++++++++++++-
> 2 files changed, 34 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/ice/ice_dcf.c b/drivers/net/ice/ice_dcf.c index
> 084f7a53db..366ff0a907 100644
> --- a/drivers/net/ice/ice_dcf.c
> +++ b/drivers/net/ice/ice_dcf.c
> @@ -593,6 +593,8 @@ ice_dcf_init_hw(struct rte_eth_dev *eth_dev, struct
> ice_dcf_hw *hw)
> struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
> int ret, size;
>
> + hw->resetting = false;
> +
> hw->avf.hw_addr = pci_dev->mem_resource[0].addr;
> hw->avf.back = hw;
>
> diff --git a/drivers/net/ice/ice_dcf_ethdev.c
> b/drivers/net/ice/ice_dcf_ethdev.c
> index 7c71a48010..171b58a748 100644
> --- a/drivers/net/ice/ice_dcf_ethdev.c
> +++ b/drivers/net/ice/ice_dcf_ethdev.c
> @@ -1025,11 +1025,43 @@ ice_dcf_tm_ops_get(struct rte_eth_dev *dev
> __rte_unused,
> return 0;
> }
>
> +static inline void
> +ice_dcf_simple_reset(struct rte_eth_dev *eth_dev, struct ice_dcf_hw
Better to rename the function to ice_dcf_reset_hw
> +*hw) {
> + ice_dcf_uninit_hw(eth_dev, hw);
> + ice_dcf_init_hw(eth_dev, hw);
> +}
> +
> +/* Check if reset has been triggered by PF */ static inline bool
> +dcf_is_reset(struct rte_eth_dev *dev) {
> + struct ice_dcf_adapter *ad = dev->data->dev_private;
> + struct iavf_hw *hw = &ad->real_hw.avf;
> +
> + return !(IAVF_READ_REG(hw, IAVF_VF_ARQLEN1) &
> + IAVF_VF_ARQLEN1_ARQENABLE_MASK);
> +}
> +
> static int
> ice_dcf_dev_reset(struct rte_eth_dev *dev) {
> + struct ice_dcf_adapter *ad = dev->data->dev_private;
> + struct ice_dcf_hw *hw = &ad->real_hw;
> int ret;
>
> + if (dcf_is_reset(dev)) {
> + if (!ad->real_hw.resetting)
> + ad->real_hw.resetting = true;
> + PMD_DRV_LOG(ERR, "The DCF has been reset by PF");
> +
> + /*
> + * Do the simplified reset to make DCF get AdminQ resource.
> + * Then the next uninit/init can clean filters successfully.
> + */
Can reword as below:
Simply reset hw to trigger an additional DCF enable/disable cycle which help to
workaround
the issue that kernel driver may not clean up resource during previous reset.
> + ice_dcf_simple_reset(dev, hw);
> + }
> +
> ret = ice_dcf_dev_uninit(dev);
> if (ret)
> return ret;
> @@ -1072,7 +1104,6 @@ ice_dcf_dev_init(struct rte_eth_dev *eth_dev) {
> struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
>
> - adapter->real_hw.resetting = false;
> eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
> eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
> eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
> --
> 2.27.0
