Re: [dpdk-dev] [PATCH v3] net/i40e: fix modify the number of qps in VF

[Jeff Guo <jia.guo@intel.com>]

hi, alvin

The processing of the queue pairs configure looks better than prior 
version, but still have some comment.

On 7/15/2020 3:28 PM, alvinx.zhang@intel.com wrote:
> From: Alvin Zhang <alvinx.zhang@intel.com>
>
> If a VF request PF to allocate more number of queue pairs, the PF will
> free the queue pairs which have been allocated and reset the VF. So,
> VF should stop to work until all the process is done. This patch modify
> the process of the request queue pairs. To improve efficiency and
> eliminate code redundancy, the promiscuous ops were also updated.
>
> Fixes: c48eb308ed13 ("net/i40e: support VF request more queues")
> Cc: stable@dpdk.org
>
> Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>
> ---
>
> V2: Update git log and modify codes according to comments.
> V3: All the code was refactored.
> ---
>   drivers/net/i40e/i40e_ethdev_vf.c | 109 +++++++++++++++++++++-----------------
>   1 file changed, 59 insertions(+), 50 deletions(-)
>
> diff --git a/drivers/net/i40e/i40e_ethdev_vf.c b/drivers/net/i40e/i40e_ethdev_vf.c
> index eca716a..758d444 100644
> --- a/drivers/net/i40e/i40e_ethdev_vf.c
> +++ b/drivers/net/i40e/i40e_ethdev_vf.c
> @@ -91,7 +91,8 @@ static int i40evf_vlan_filter_set(struct rte_eth_dev *dev,
>   				  uint16_t vlan_id, int on);
>   static int i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
>   static void i40evf_dev_close(struct rte_eth_dev *dev);
> -static int  i40evf_dev_reset(struct rte_eth_dev *dev);
> +static int i40evf_dev_reset(struct rte_eth_dev *dev);
> +static int i40evf_check_vf_reset_done(struct rte_eth_dev *dev);
>   static int i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev);
>   static int i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev);
>   static int i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev);
> @@ -519,10 +520,17 @@ struct rte_i40evf_xstats_name_off {
>   
>   	err = i40evf_execute_vf_cmd(dev, &args);
>   
> -	if (err)
> -		PMD_DRV_LOG(ERR, "fail to execute command "
> -			    "CONFIG_PROMISCUOUS_MODE");
> -	return err;
> +	if (err) {
> +		if (err == I40E_NOT_SUPPORTED)
> +			return -ENOTSUP;
> +
> +		PMD_DRV_LOG(ERR, "Failed to set promiscuous mode");
> +		return -EAGAIN;


I think below is better, please try to keep the log align in driver if 
there is no special reason.

	if (err) {
		PMD_DRV_LOG(ERR, "fail to execute command "
			    "CONFIG_PROMISCUOUS_MODE");
  
		if (err == I40E_NOT_SUPPORTED)
			return -ENOTSUP;
		else
			return -EAGAIN;

> +	}
> +
> +	vf->promisc_unicast_enabled = enable_unicast;
> +	vf->promisc_multicast_enabled = enable_multicast;
> +	return 0;
>   }
>   
>   static int
> @@ -1081,12 +1089,31 @@ static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
>   	args.out_size = I40E_AQ_BUF_SZ;
>   
>   	rte_eal_alarm_cancel(i40evf_dev_alarm_handler, dev);
> +
>   	err = i40evf_execute_vf_cmd(dev, &args);
> -	if (err)
> +	if (err != I40E_SUCCESS) {
> +		rte_eal_alarm_set(I40EVF_ALARM_INTERVAL,
> +				  i40evf_dev_alarm_handler, dev);
>   		PMD_DRV_LOG(ERR, "fail to execute command OP_REQUEST_QUEUES");
> +		return -EIO;
> +	}
>   
>   	rte_eal_alarm_set(I40EVF_ALARM_INTERVAL,
>   			  i40evf_dev_alarm_handler, dev);
> +
> +	/*
> +	 * The PF will issue a reset to the VF when change the number of
> +	 * queues. The PF will set I40E_VFGEN_RSTAT to COMPLETE first, then
> +	 * wait 10ms and set it to ACTIVE. In this duration, vf may not catch
> +	 * the moment that COMPLETE is set. So, for vf, we'll try to wait a
> +	 * long time.
> +	 */
> +	rte_delay_ms(100);
> +
> +	err = i40evf_check_vf_reset_done(dev);
> +	if (err)
> +		PMD_DRV_LOG(ERR, "VF is still resetting");
> +


Please keep the return value and eliminate the duplicate, do you think 
below is better, you could ref

  	err = i40evf_execute_vf_cmd(dev, &args);

  	rte_eal_alarm_set(I40EVF_ALARM_INTERVAL,
  			  i40evf_dev_alarm_handler, dev);

	if (!err) {
		check vf reset done....
	} else {
		PMD_DRV_LOG(ERR, "fail to execute command OP_REQUEST_QUEUES");
	}

>   	return err;
>   }
>   
> @@ -1514,7 +1541,7 @@ static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
>   	hw->bus.device = pci_dev->addr.devid;
>   	hw->bus.func = pci_dev->addr.function;
>   	hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
> -	hw->adapter_stopped = 0;
> +	hw->adapter_stopped = 1;
>   	hw->adapter_closed = 0;
>   
>   	/* Pass the information to the rte_eth_dev_close() that it should also
> @@ -1610,10 +1637,27 @@ static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
>   	ad->tx_vec_allowed = true;
>   
>   	if (num_queue_pairs > vf->vsi_res->num_queue_pairs) {
> -		int ret = 0;
> +		struct i40e_hw *hw;
> +		int ret;
> +
> +		/*
> +		 * All VF resources will be reallocated, so change queue pairs
> +		 * in secondary processes is forbidden.
> +		 */


Please don't use an empty /* line, use /* Comment....

And seems that the word "forbidden" is not very make sense. you could 
ref other place about secondary process checking, simply like below

             / * for secondary processes, we don't configure queue pairs 
any further

               * as primary has already done this work.

              */


> +		if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
> +			PMD_DRV_LOG(ERR,
> +				"For secondary processes, change queue pairs is forbidden!");


Alignment should match open parenthesis, you could use the checkpatch.pl 
to check it.

Add please check if below i40evf_init_vlan also no need in secondary 
process. If so, i suggest split this part into other specific fixing 
patch for secondary process configuration.


> +			return -ENOTSUP;
> +		}
>   
> +		hw  = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
>   		PMD_DRV_LOG(INFO, "change queue pairs from %u to %u",
>   			    vf->vsi_res->num_queue_pairs, num_queue_pairs);
> +		if (hw->adapter_stopped == 0) {
> +			PMD_DRV_LOG(WARNING, "Device must be stopped first!");
> +			return -EINVAL;


I think ERR but not warning should be return here, and do you think -EBUSY should be better than -EINVAL?


> +		}
> +
>   		ret = i40evf_request_queues(dev, num_queue_pairs);
>   		if (ret != 0)
>   			return ret;
> @@ -2182,68 +2226,32 @@ static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
>   i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev)
>   {
>   	struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
> -	int ret;
> -
> -	ret = i40evf_config_promisc(dev, 1, vf->promisc_multicast_enabled);
> -	if (ret == 0)
> -		vf->promisc_unicast_enabled = TRUE;
> -	else if (ret == I40E_NOT_SUPPORTED)
> -		ret = -ENOTSUP;
> -	else
> -		ret = -EAGAIN;
>   
> -	return ret;
> +	return i40evf_config_promisc(dev, true, vf->promisc_multicast_enabled);
>   }
>   
>   static int
>   i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev)
>   {
>   	struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
> -	int ret;
> -
> -	ret = i40evf_config_promisc(dev, 0, vf->promisc_multicast_enabled);
> -	if (ret == 0)
> -		vf->promisc_unicast_enabled = FALSE;
> -	else if (ret == I40E_NOT_SUPPORTED)
> -		ret = -ENOTSUP;
> -	else
> -		ret = -EAGAIN;
>   
> -	return ret;
> +	return i40evf_config_promisc(dev, false, vf->promisc_multicast_enabled);
>   }
>   
>   static int
>   i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev)
>   {
>   	struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
> -	int ret;
> -
> -	ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 1);
> -	if (ret == 0)
> -		vf->promisc_multicast_enabled = TRUE;
> -	else if (ret == I40E_NOT_SUPPORTED)
> -		ret = -ENOTSUP;
> -	else
> -		ret = -EAGAIN;
>   
> -	return ret;
> +	return i40evf_config_promisc(dev, vf->promisc_unicast_enabled, true);
>   }
>   
>   static int
>   i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev)
>   {
>   	struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
> -	int ret;
> -
> -	ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 0);
> -	if (ret == 0)
> -		vf->promisc_multicast_enabled = FALSE;
> -	else if (ret == I40E_NOT_SUPPORTED)
> -		ret = -ENOTSUP;
> -	else
> -		ret = -EAGAIN;
>   
> -	return ret;
> +	return i40evf_config_promisc(dev, vf->promisc_unicast_enabled, false);
>   }
>   
>   static int
> @@ -2365,8 +2373,9 @@ static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
>   	 * it is a workaround solution when work with kernel driver
>   	 * and it is not the normal way
>   	 */
> -	i40evf_dev_promiscuous_disable(dev);
> -	i40evf_dev_allmulticast_disable(dev);
> +	if (vf->promisc_unicast_enabled || vf->promisc_multicast_enabled)
> +		i40evf_config_promisc(dev, false, false);


Need checking the return status when i40evf_config_promisc return failed?


> +
>   	rte_eal_alarm_cancel(i40evf_dev_alarm_handler, dev);
>   
>   	i40evf_reset_vf(dev);