>David Ahern <dsahern@xxxxxxxxx> writes: > On 7/14/20 6:29 AM, Toke Høiland-Jørgensen wrote: >> Hangbin Liu <liuhangbin@xxxxxxxxx> writes: >> >>> This patch is for xdp multicast support. which has been discussed before[0], >>> The goal is to be able to implement an OVS-like data plane in XDP, i.e., >>> a software switch that can forward XDP frames to multiple ports. >>> >>> To achieve this, an application needs to specify a group of interfaces >>> to forward a packet to. It is also common to want to exclude one or more >>> physical interfaces from the forwarding operation - e.g., to forward a >>> packet to all interfaces in the multicast group except the interface it >>> arrived on. While this could be done simply by adding more groups, this >>> quickly leads to a combinatorial explosion in the number of groups an >>> application has to maintain. >>> >>> To avoid the combinatorial explosion, we propose to include the ability >>> to specify an "exclude group" as part of the forwarding operation. This >>> needs to be a group (instead of just a single port index), because there >>> may have multi interfaces you want to exclude. >>> >>> Thus, the logical forwarding operation becomes a "set difference" >>> operation, i.e. "forward to all ports in group A that are not also in >>> group B". This series implements such an operation using device maps to >>> represent the groups. This means that the XDP program specifies two >>> device maps, one containing the list of netdevs to redirect to, and the >>> other containing the exclude list. >>> >>> To achieve this, I re-implement a new helper bpf_redirect_map_multi() >>> to accept two maps, the forwarding map and exclude map. If user >>> don't want to use exclude map and just want simply stop redirecting back >>> to ingress device, they can use flag BPF_F_EXCLUDE_INGRESS. >>> >>> The 2nd and 3rd patches are for usage sample and testing purpose, so there >>> is no effort has been made on performance optimisation. I did same tests >>> with pktgen(pkt size 64) to compire with xdp_redirect_map(). Here is the >>> test result(the veth peer has a dummy xdp program with XDP_DROP directly): >>> >>> Version | Test | Native | Generic >>> 5.8 rc1 | xdp_redirect_map i40e->i40e | 10.0M | 1.9M >>> 5.8 rc1 | xdp_redirect_map i40e->veth | 12.7M | 1.6M >>> 5.8 rc1 + patch | xdp_redirect_map i40e->i40e | 10.0M | 1.9M >>> 5.8 rc1 + patch | xdp_redirect_map i40e->veth | 12.3M | 1.6M >>> 5.8 rc1 + patch | xdp_redirect_map_multi i40e->i40e | 7.2M | 1.5M >>> 5.8 rc1 + patch | xdp_redirect_map_multi i40e->veth | 8.5M | 1.3M >>> 5.8 rc1 + patch | xdp_redirect_map_multi i40e->i40e+veth | 3.0M | 0.98M >>> >>> The bpf_redirect_map_multi() is slower than bpf_redirect_map() as we loop >>> the arrays and do clone skb/xdpf. The native path is slower than generic >>> path as we send skbs by pktgen. So the result looks reasonable. >>> >>> Last but not least, thanks a lot to Jiri, Eelco, Toke and Jesper for >>> suggestions and help on implementation. >>> >>> [0] https://xdp-project.net/#Handling-multicast >>> >>> v7: Fix helper flag check >>> Limit the *ex_map* to use DEVMAP_HASH only and update function >>> dev_in_exclude_map() to get better performance. >> >> Did it help? The performance numbers in the table above are the same as >> in v6... >> > > If there is only 1 entry in the exclude map, then the numbers should be > about the same. I would still expect the lack of the calls to devmap_get_next_key() to at least provide a small speedup, no? That the numbers are completely unchanged looks a bit suspicious... -Toke
