On Mon, Sep 12, 2022 at 4:47 PM Flavio Leitner <[email protected]> wrote:
> On Mon, Sep 12, 2022 at 5:52 AM David Marchand <[email protected]> 
> wrote:
>> On Fri, Sep 9, 2022 at 7:58 PM Flavio Leitner <[email protected]> wrote:
>> >
>> > Thanks for working on this patch!
>> >
>> > It seems possible to change this patch later when the other TSO series
>> > gets merged to disable TSO only on the affected port.
>> > Mike, any thoughts?
>>
>> It should be what this patch already does: disable TSO only for the
>> affected port as I mentionned in the commitlog below.
>> Can you clarify?
>
>
> Oops, I should have added more context. The issue is that TSO is 
> all-or-nothing
> because there is no software fallback currently [see 
> netdev_send_prepare_packet()].
> So, if the problem happens and the reconnection with disabled TSO succeeds, 
> all
> TSO packets to that port will be dropped. The question is "should we allow 
> the update
> but run the risk of dropping packets or fail to update?"
>
> The software fallback  is provided in the posted TSO series. In that case, we 
> can
> have a per-port TSO knob because if it gets disabled, the packets will be 
> segmented
> in software and not dropped.
> So, the patch makes sense with the posted TSO series applied, but I am not so 
> sure
> with the current code.
>
> What do you think?

Ok, I get your point now.

Btw, I still have some review to finish on the TSO series.
For the time being, I will mark my patch as "Changes requested",
rebase it and test on top of TSO series.


-- 
David Marchand

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to