[dpdk-dev] [PATCH v1 0/3] ethdev: document more retained across restart

Andrew Rybchenko arybchenko at solarflare.com
Wed Jan 23 13:23:56 CET 2019


On 1/10/19 5:03 AM, WILLIAMS, CHARLES J wrote:
>
>
> On 1/9/19 2:15 AM, Andrew Rybchenko wrote:
>> On 1/8/19 5:52 PM, Ferruh Yigit wrote:
>>> On 12/19/2018 6:22 AM, Andrew Rybchenko wrote:
>>>> The patch series tries to improve documentation of what is retained
>>>> across default restart.
>>> Overall makes sense to add below items into retained list, only concern if is
>>> there any PMD conflicts with these information, they should either updated with
>>> this patch or at least notified about expectation change.
>>
>> From my point of view it is just clarification of the required behaviour.
>> MTU is required because of flag which may be used to advertise that
>> it is impossible to change in started state (otherwise the behaviour for
>> different PMDs will be absolutely different).
>> Default MAC is just cosmetics because of MAC address list is already
>> mentioned and the only goal is to highlight since these are different
>> features from the feature list point of view.
>> All-multicast is a part of Rx mode.
>>
>> In fact I recall that net/bonding does not preserve all-multicast
>> (CC maintainers). If there is an agreement to fix it, I can take
>> a look - it should not be hard to fix.
>
> I don't think bonding does anything particular because the 
> all_multicast state of the bonding PMD isn't related to the slaves.  
> Currently, it's up to the applications to correctly configure 
> multicast on the slaves.
>

May be I'm wrong but I think all-multicast handling should be similar to 
promiscuous mode handling in bonding: for all slaves in the case of 
round-robin, balance and broadcast, primary only in the case of active 
backup, TLB and ALB and when slave is added/removed in 802.3ad case.

> On a side note, I don't think the registered multicast addresses are 
> preserved across PMD stop/start (unless this has been fixed recently).
>

Yes, that's true, but I think it is separate story.

>>
>> In general I think that notification is sufficient in this case.
>>
>>>> Andrew Rybchenko (3):
>>>>    ethdev: advertise MTU as retained across device stop/start
>>>>    ethdev: advertise default MAC as retained on device restart
>>>>    ethdev: highlight that all-multicast is retained on restart
>>>>
>>>>   lib/librte_ethdev/rte_ethdev.h | 6 ++++--
>>>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>>>
>>



More information about the dev mailing list