[dpdk-dev] [PATCH V4 8/9] app/testpmd: show example to handle hot unplug

Guo, Jia jia.guo at intel.com
Thu Jul 5 09:54:15 CEST 2018



On 7/4/2018 3:06 PM, Matan Azrad wrote:
> Hi Thomas, Guo
>
> From: Thomas Monjalon
>> 03/07/2018 11:35, Guo, Jia:
>>> On 7/1/2018 3:46 PM, Matan Azrad wrote:
>>>> From: Jeff Guo
>>>>> --- a/app/test-pmd/testpmd.c
>>>>> +++ b/app/test-pmd/testpmd.c
>>>>> @@ -2206,9 +2209,12 @@ eth_dev_event_callback(char
>> *device_name,
>>>>> enum rte_dev_event_type type,
>>>>>    	case RTE_DEV_EVENT_REMOVE:
>>>>>    		RTE_LOG(ERR, EAL, "The device: %s has been removed!\n",
>>>>>    			device_name);
>>>>> -		/* TODO: After finish failure handle, begin to stop
>>>>> -		 * packet forward, stop port, close port, detach port.
>>>>> -		 */
>>>>> +		ret = rte_eth_dev_get_port_by_name(device_name,
>> &port_id);
>>>> As you probably know, 1 rte_device may be associated to more than one
>> ethdev ports, so the ethdev port name can be different from rte_device
>> name.
>>>> Looks like we need a new ethdev API to get all the ports associated to
>> one rte_device.
>>> agree, seems that the the old ethdev API have some issue when got all
>>> port by device name. we could check with ethdev maintainer and fix it
>>> by specific ethdev patch later.
>> This ethdev function could return an error if several ports match.
>>
> Just to clarify:
>
> The  ethdev name may be different from the rte_device name of a port,
> The rte_eth_dev_get_port_by_name() searches the ethdev name and not the rte_device name.
>
>> Ideally, we should not use this function at all.
>> If you want to manage an ethdev port, why are you using an EAL event?
>> There is an ethdev callback mechanism for port removal.
> So, looks like the EAL event should trigger an ethdev event for all the ports associated to this rte_device.
> I think that the best one to do it is the PMD, so maybe the PMD(which wants to support hot unplug) should register to the EAL event and to trigger an ethdev RMV event from the EAL callback.
>
> What do you think?
>

i think matan give an constructive option to combine the usage of eal 
event and ethdev event,  but i am not sure which is the best one.
So let this discuss on going, i will remove this 8/9 and 9/9 patches, 
let the patch set focus on the hotplug failure mechanism ,
and will use another patch set to cover the event management example in 
testpmd.

>   
>



More information about the dev mailing list