[ovs-dev] [PATCH 6/9] netdev: Add support multiqueue recv.

Pravin Shelar pshelar at nicira.com
Wed Mar 19 22:20:53 UTC 2014


On Wed, Mar 19, 2014 at 11:26 AM, Ben Pfaff <blp at nicira.com> wrote:
> On Tue, Mar 18, 2014 at 01:53:52PM -0700, Pravin wrote:
>> new netdev type like DPDK can support multi-queue IO. Following
>> patch Adds support for same.
>>
>> Signed-off-by: Pravin B Shelar <pshelar at nicira.com>
>
> Needs comment and documentation updates.
>
> In particular, from reading the code I infer (but I'm not confident)
> that now a device that doesn't support receiving packets would now
> return an nr_rx of 0, and that the netdev layer doesn't really yet
> support devices with more than one queue, since I don't see a way to
> get a netdev->nr_rx greater than 1.  Maybe the intent is that the
> netdev's class->construct() sets nr_rx if it has more than one queue?
> If so then please document that in netdev-provider.h's comment for
> construct().
>
yes, construct is support to change default value. I will document it.

> Maybe netdev_rxq_open() should reject calls for which id >=
> netdev->nr_rx, so that implementations don't have to (I think that all
> the implementations are technically broken as of this commit).
>
ok, I will add chack in netdev_rxq_open(). Why do you think
implementations are broken?

> We commonly use an n_ prefix for number in userspace.  I don't think
> we have many examples of nr_.
>
ok.

> I think that the two iterations through the hmap of ports in
> pmd_load_queues() could be just one if we just used a function like
> x2nrealloc() to allocate the port_list as we go.  In the end, it's
> probably easier to understand to do it that way, and I doubt this code
> is time critical.  (I guess that should have been a comment on an
> earlier patch, sorry.)
>
ok, x2nrealloc() is good idea.



More information about the dev mailing list