| Message ID | 20210226112322.144927-1-bjorn.topel@gmail.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Optimize bpf_redirect_map()/xdp_do_redirect() | expand |
Björn Töpel <bjorn.topel@gmail.com> writes: > Hi XDP-folks, > > This two patch series contain two optimizations for the > bpf_redirect_map() helper and the xdp_do_redirect() function. > > The bpf_redirect_map() optimization is about avoiding the map lookup > dispatching. Instead of having a switch-statement and selecting the > correct lookup function, we let bpf_redirect_map() be a map operation, > where each map has its own bpf_redirect_map() implementation. This way > the run-time lookup is avoided. > > The xdp_do_redirect() patch restructures the code, so that the map > pointer indirection can be avoided. > > Performance-wise I got 3% improvement for XSKMAP > (sample:xdpsock/rx-drop), and 4% (sample:xdp_redirect_map) on my > machine. > > More details in each commit. > > @Jesper/Toke I dropped your Acked-by: on the first patch, since there > were major restucturing. Please have another look! Thanks! Will do! Did you update the performance numbers above after that change? -Toke
On 2021-02-26 12:35, Toke Høiland-Jørgensen wrote: > Björn Töpel <bjorn.topel@gmail.com> writes: > >> Hi XDP-folks, >> >> This two patch series contain two optimizations for the >> bpf_redirect_map() helper and the xdp_do_redirect() function. >> >> The bpf_redirect_map() optimization is about avoiding the map lookup >> dispatching. Instead of having a switch-statement and selecting the >> correct lookup function, we let bpf_redirect_map() be a map operation, >> where each map has its own bpf_redirect_map() implementation. This way >> the run-time lookup is avoided. >> >> The xdp_do_redirect() patch restructures the code, so that the map >> pointer indirection can be avoided. >> >> Performance-wise I got 3% improvement for XSKMAP >> (sample:xdpsock/rx-drop), and 4% (sample:xdp_redirect_map) on my >> machine. >> >> More details in each commit. >> >> @Jesper/Toke I dropped your Acked-by: on the first patch, since there >> were major restucturing. Please have another look! Thanks! > > Will do! Did you update the performance numbers above after that change? > I did. The XSKMAP performance stayed the same (no surprise, since the code was the same). However, for the DEVMAP the v4 got rid of a call, so it *should* be a bit better, but for some reason it didn't show on my machine. Björn
Björn Töpel <bjorn.topel@intel.com> writes: > On 2021-02-26 12:35, Toke Høiland-Jørgensen wrote: >> Björn Töpel <bjorn.topel@gmail.com> writes: >> >>> Hi XDP-folks, >>> >>> This two patch series contain two optimizations for the >>> bpf_redirect_map() helper and the xdp_do_redirect() function. >>> >>> The bpf_redirect_map() optimization is about avoiding the map lookup >>> dispatching. Instead of having a switch-statement and selecting the >>> correct lookup function, we let bpf_redirect_map() be a map operation, >>> where each map has its own bpf_redirect_map() implementation. This way >>> the run-time lookup is avoided. >>> >>> The xdp_do_redirect() patch restructures the code, so that the map >>> pointer indirection can be avoided. >>> >>> Performance-wise I got 3% improvement for XSKMAP >>> (sample:xdpsock/rx-drop), and 4% (sample:xdp_redirect_map) on my >>> machine. >>> >>> More details in each commit. >>> >>> @Jesper/Toke I dropped your Acked-by: on the first patch, since there >>> were major restucturing. Please have another look! Thanks! >> >> Will do! Did you update the performance numbers above after that change? >> > > I did. The XSKMAP performance stayed the same (no surprise, since the > code was the same). However, for the DEVMAP the v4 got rid of a call, so > it *should* be a bit better, but for some reason it didn't show on my > machine. Alright, fair enough - pesky real world not lining up with expectations! Maybe Jesper has additional suggestions, but I can live with the 4% improvement ;) -Toke
Hi XDP-folks, This two patch series contain two optimizations for the bpf_redirect_map() helper and the xdp_do_redirect() function. The bpf_redirect_map() optimization is about avoiding the map lookup dispatching. Instead of having a switch-statement and selecting the correct lookup function, we let bpf_redirect_map() be a map operation, where each map has its own bpf_redirect_map() implementation. This way the run-time lookup is avoided. The xdp_do_redirect() patch restructures the code, so that the map pointer indirection can be avoided. Performance-wise I got 3% improvement for XSKMAP (sample:xdpsock/rx-drop), and 4% (sample:xdp_redirect_map) on my machine. More details in each commit. @Jesper/Toke I dropped your Acked-by: on the first patch, since there were major restucturing. Please have another look! Thanks! Changelog: v3->v4: Made bpf_redirect_map() a map operation. (Daniel) v2->v3: Fix build when CONFIG_NET is not set. (lkp) v1->v2: Removed warning when CONFIG_BPF_SYSCALL was not set. (lkp) Cleaned up case-clause in xdp_do_generic_redirect_map(). (Toke) Re-added comment. (Toke) rfc->v1: Use map_id, and remove bpf_clear_redirect_map(). (Toke) Get rid of the macro and use __always_inline. (Jesper) rfc: https://lore.kernel.org/bpf/87im7fy9nc.fsf@toke.dk/ (Cover not on lore!) v1: https://lore.kernel.org/bpf/20210219145922.63655-1-bjorn.topel@gmail.com/ v2: https://lore.kernel.org/bpf/20210220153056.111968-1-bjorn.topel@gmail.com/ v3: https://lore.kernel.org/bpf/20210221200954.164125-3-bjorn.topel@gmail.com/ Cheers, Björn Björn Töpel (2): bpf, xdp: make bpf_redirect_map() a map operation bpf, xdp: restructure redirect actions include/linux/bpf.h | 26 ++---- include/linux/filter.h | 39 +++++++- include/net/xdp_sock.h | 19 ---- include/trace/events/xdp.h | 66 ++++++++----- kernel/bpf/cpumap.c | 10 +- kernel/bpf/devmap.c | 19 +++- kernel/bpf/verifier.c | 11 ++- net/core/filter.c | 183 ++++++++++++------------------------- net/xdp/xskmap.c | 20 +++- 9 files changed, 195 insertions(+), 198 deletions(-) base-commit: 9c8f21e6f8856a96634e542a58ef3abf27486801