Re: [dpdk-dev] [PATCH 5/7] ethdev: do nothing if all-multicast mode is applied again

[Andrew Rybchenko <arybchenko@solarflare.com>]

On 9/24/19 2:03 PM, Ferruh Yigit wrote:
> On 9/24/2019 9:54 AM, Andrew Rybchenko wrote:
>> On 9/24/19 11:36 AM, Ferruh Yigit wrote:
>>> On 9/9/2019 1:13 PM, Andrew Rybchenko wrote:
>>>> Since driver callbacks return status code now, there is no necessity
>>>> to enable or disable all-multicast mode once again if it is already
>>>> successfully enabled or disabled.
>>>>
>>>> Configuration restore at startup tries to ensure that configured
>>>> all-multicast mode is applied and start will return error if it fails.
>>>>
>>>> Also it avoids theoretical cases when already configured all-multicast
>>>> mode is applied once again and fails. In this cases it is unclear
>>>> which value should be reported on get (configured or opposite).
>>>>
>>>> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
>>>> ---
>>>>    lib/librte_ethdev/rte_ethdev.c | 40 ++++++++++++++++++++--------------
>>>>    1 file changed, 24 insertions(+), 16 deletions(-)
>>>>
>>>> diff --git a/lib/librte_ethdev/rte_ethdev.c b/lib/librte_ethdev/rte_ethdev.c
>>>> index 8115226c91..e1921e8225 100644
>>>> --- a/lib/librte_ethdev/rte_ethdev.c
>>>> +++ b/lib/librte_ethdev/rte_ethdev.c
>>>> @@ -1416,16 +1416,22 @@ rte_eth_dev_config_restore(struct rte_eth_dev *dev,
>>>>    	}
>>>>    
>>>>    	/* replay all multicast configuration */
>>>> -	if (rte_eth_allmulticast_get(port_id) == 1) {
>>>> -		ret = rte_eth_allmulticast_enable(port_id);
>>>> +	/*
>>>> +	 * use callbacks directly since we don't need port_id check and
>>>> +	 * would like to bypass the same value set
>>>> +	 */
>>>> +	if (rte_eth_allmulticast_get(port_id) == 1 &&
>>>> +	    *dev->dev_ops->allmulticast_enable != NULL) {
>>>> +		ret = (*dev->dev_ops->allmulticast_enable)(dev);
>>> I am for using the API here, it is more abstract instead of adding the dev_ops
>>> null checks etc. Will there be any downside to use the API?
>> API does not call operation if value matches and it will exactly match here
>> for sure since we just get it and applying once again.
> Ah, right, we need it here as you explained. Thx.

In fact, I think eth_err() is required to handle callback return value.
I'll add it in v2.

>> Can't say that I like usage callback directly here, but it looks acceptable
>> for me. I've tried to clarify why it is done this way in the comment above.
>>
>>> <...>
>>>
>>>> @@ -1962,16 +1968,17 @@ int
>>>>    rte_eth_allmulticast_enable(uint16_t port_id)
>>>>    {
>>>>    	struct rte_eth_dev *dev;
>>>> -	uint8_t old_allmulticast;
>>>> -	int diag;
>>>> +	int diag = 0;
>>>>    
>>>>    	RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>>>>    	dev = &rte_eth_devices[port_id];
>>>>    
>>>>    	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->allmulticast_enable, -ENOTSUP);
>>>> -	old_allmulticast = dev->data->all_multicast;
>>>> -	diag = (*dev->dev_ops->allmulticast_enable)(dev);
>>>> -	dev->data->all_multicast = (diag == 0) ? 1 : old_allmulticast;
>>>> +
>>>> +	if (dev->data->all_multicast == 0) {
>>> What about adding this check even before 'allmulticast_enable' check, so if the
>>> multicast is already enabled why bother having dev_ops or not:
>>>
>>> if (dev->data->all_multicast == 1)
>>> 	return eth_err(port_id, diag);
>> Yes, it is a good idea. If so, similar fix up will be required for
>> promiscuous mode.

I'd prefer to return 0 directly.

>>>> +		diag = (*dev->dev_ops->allmulticast_enable)(dev);
>>>> +		dev->data->all_multicast = (diag == 0) ? 1 : 0;
>>>> +	}
>>>>    
>>>>    	return eth_err(port_id, diag);
>>>>    }
>>>> @@ -1980,18 +1987,19 @@ int
>>>>    rte_eth_allmulticast_disable(uint16_t port_id)
>>>>    {
>>>>    	struct rte_eth_dev *dev;
>>>> -	uint8_t old_allmulticast;
>>>> -	int diag;
>>>> +	int diag = 0;
>>>>    
>>>>    	RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>>>>    	dev = &rte_eth_devices[port_id];
>>>>    
>>>>    	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->allmulticast_disable, -ENOTSUP);
>>>> -	old_allmulticast = dev->data->all_multicast;
>>>> -	dev->data->all_multicast = 0;
>>>> -	diag = (*dev->dev_ops->allmulticast_disable)(dev);
>>>> -	if (diag != 0)
>>>> -		dev->data->all_multicast = old_allmulticast;
>>>> +
>>>> +	if (dev->data->all_multicast == 1) {
>>> Same comment with above, can we move this check above..
>> Yes, will fix in the next version as well. Thanks for review.
>>
>>>> +		dev->data->all_multicast = 0;
>>>> +		diag = (*dev->dev_ops->allmulticast_disable)(dev);
>>>> +		if (diag != 0)
>>>> +			dev->data->all_multicast = 1;
>>>> +	}
>>>>    
>>>>    	return eth_err(port_id, diag);
>>>>    }
>>>>