[ovs-dev] [PATCH V2 1/3] dpif-netdev: Do not flush PMD offloads on reload

[Ilya Maximets]

On 7/25/21 8:40 AM, Eli Britstein wrote:
> 
> On 7/23/2021 9:00 PM, Ilya Maximets wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> On 7/12/21 5:07 PM, Eli Britstein wrote:
>>> Before flushing offloads of a removed port was supported by [1], it was
>>> necessary to flush the 'marks'. In doing so, all offloads of the PMD are
>>> removed, include the ones that are not related to the removed port and
>>> that are not modified following this removal. As a result such flows are
>>> evicted from being offloaded, and won't resume offloading.
>>>
>>> As PMD offload flush is not necessary, avoid it.
>>>
>>> [1] 62d1c28e9ce0 ("dpif-netdev: Flush offload rules upon port deletion.")
>>>
>>> Signed-off-by: Eli Britstein <elibr at nvidia.com>
>>> Reviewed-by: Gaetan Rivet <gaetanr at nvidia.com>
>>> ---
>> Is my understanding here correct:
>> On a port deletion netdev_fow_flush() will remove flows from HW and
>> offloading layer.  Later, ofproto will request to remove flows from
>> the datapath.  flow marks will be freed, but actual netdev_flow_del()
>> will fail, because netdev-offload already removed these flows.  But
>> we do not really care about this failure.  Right?
> 
> That is correct.
> 
> It was also the same before above [1], but then in the race condition, the offload memory was leaked, and the offloads themselves were either not destroyed (e.g. leaked) or removed by the PMD, which was PMD dependent.
> 
> [1] resolved the leaking, but didn't change the mentioned failure behavior.
> 
> This commit doesn't prevents this failure either, but removes the code that destroys offloads of flows that should not be destroyed.

Make sense.  Thanks!