On Fri, May 18, 2018 at 01:05:52AM +0300, Leon Romanovsky wrote: > > I think rdmatool might be new enough, IP_GIDS rare enough, and the > > defect harmless enough that we can fix it properly. > > > > Netlink should report the raw IB port capability mask in that > > attribute, not some data from the uverbs interface. > > > > IMHO, just drop IP_GIDS from iproute2 and we are good. > > RDMtool is part of my current distribution and deployment of mlx4/mlx5 > is so large, so I can't agree with you that it is "rare enough". > > I'll review your change on Monday, but I have gut feeling that you broke > backward compatibility with already deployed rdmatool. It isn't so simple. Today if you run rdmatool against qib (and probably mlx devices in IB mode) it will reqport IP_GIDS, but that is wrong, it is actually IsOtherLocalChangesNoticeSupported, and it will print a bunch of UNKNOWNS because it doesn't decode all the bits the kernel potentially sets. So it is already broken today.. The sane fix is to restore the netlink attribute to what it says it is (the port capability mask), and fix iptool to print all 31 bits the kernel supports with their proper labels. IP_GIDs itself is very obscure and I can't belive anyone is going to be relying on a half working 8 month old rdmatool feature at this point. So lets fix it sensibly. Jason -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
