On 16-08-13 09:11 AM, Richard Cochran wrote: > On Sat, Aug 13, 2016 at 08:27:38AM -0700, Alexander Duyck wrote: >> I really don' think this patch is going to work. If you are going to >> implement something like this and have a hope to get it accepted into >> the Linux kernel you need to come up with a solution that will work >> fore more than this one device. We don't want the drivers having to >> carry around their own sysfs controls for things that really are not >> proprietary to the device. There needs to be a generic kernel >> interface for this. The fact is something like QAV more than likely >> exists on other devices as well so it may be worth while to look into >> seeing if you could come up with some way of interfacing this with >> either ethtool ,iproute2, or maybe even the DCB/LLDP utilities since >> this is essentially splitting the Tx into two separate traffic >> classes. > > Yes to all of this. > >> Also for these kind of patches it would be best to include the netdev >> mailing list. That way it can be reviewed by a wider audience and you >> are much more likely to get this accepted upstream rather than have it >> rejected when Jeff Kirsher attempts to submit it. > > Right. We just had a discussion about implementing TSN, and we will > need proper infrastructure in place *before* we start hacking > drivers. > > Thanks, > Richard
Ah reading my email backwards. I think we could add TSN under the mqprio qdisc and ./net/dcb infrastructure. In hindsight I wouldn't have named the infrastructure dcb as its already being used for other 802.1Q things and hardware scheduling algorithms that are not strictly DCB. .John
