Date
1 - 2 of 2
minutes: IO Visor TSC/Dev Meeting
|
Brenden Blanco
Hi All,
Thanks for attending the meeting today. As usual, here are my notes. === Discussion === Bjorn - AF_XDP zero copy implementation RFC on netdev - some uapi concerns, may need to back out changes - but, over next couple weeks before merge window, discuss and improve in hopes of merging - clean up non-intel code, respin, leave out i40e patches at the end - separate i40e cleanup spin - bcm may try to join as 2nd vendor to flesh out uapi - some discussion of common code organization - xdp in separate file helps backporting Yonghong - some python3 fixes - proposals for improved container support - rewriter complexity increasing - tracing event introspection Martin - BTF id landed - working on bpftool support - still using pahole to convert dwarf - still unclear how integration with llvm will look Jiong - dominator tree info RFCs pushed - significant mem/exec time costs to run (e.g. 2x more, 3x slower) John - induction variable tracking in progress - next: min/max value tracking both: to work on topic branch under bpf-next William - tried workaround from Yonghong - still need to dig into root cause Joe - implementation of sock lookup going through iterations - concerns over locking in bpf being a leaky implementation detail === Attendees === Brenden Blanco Alexei Starovoitov Bjorn Topel Daniel Borkmann Jesper Brouer Jiong Wang Magnus Karlsson Quentin Monnet Joe Stringer Andy Gospodarek Ferris Brendan Gregg Jakub Kicinski William Tu Yonghong Song Alexander Duyck Sandipan John F |
|
|
|
Jiong Wang
On 17/05/2018 05:38, Brenden Blanco via iovisor-dev wrote:
Hi All,Just for the record, if I listened correctly, John proposed to introduce a switch for the new CFG infrastructure, i.e, we keep the exisiting DFS based check_cfg, the new CFG code will only be enabled when check_cfg detected a loop and wants know whether it is bounded, by this we could avoid extra mem/exec time cost on programs that don't have loop. Regards, Jiong John |
|
|