On 11/2/2023 4:52 PM, Stephen Hemminger wrote: > On Thu, 2 Nov 2023 15:11:10 +0000 > Ferruh Yigit <ferruh.yi...@amd.com> wrote: > >>> Stephen Hemminger (2): >>> net/tap: support infrastructure to build the BPF filter >>> net/tap; rebuild and update the BPF flow program >>> >> >> Thanks Stephen for fixing this. >> >> >> But considering it was broken for a while and nobody complained, and >> initial developers seems lost interest and it is relatively hard code to >> maintain, do we still want to keep this support. >> >> Can we probe again motivation and benefit of eBPF support in tap PMD, >> and if it is still relevant? > > > The use case was allowing an rte_flow match to a set of queues and have > the BPF program do RSS across set of rte_flow queues. > Simple non-rte_flow usage of TAP doesn't need this in most cases. > The kernel will do RSS already to multi-queue tap. > > The motivation was to allow use of rte_flow in the failsafe/tap/mlx model. > This is the legacy model for use in Hyper-V/Azure. Not aware of any > application using this. > The bug fix came from Oracle, perhaps they have more context. > > This fix set came from looking at old unmerged but ok patches in patchwork. >
Thanks for clarification. I am for getting this set and at least fix tap ebpf for this release. But does it make sense to discuss deprecation of this feature, if there is no more usecase anymore perhaps we can plan removing it?
