| Message ID | 20241004165518.120567-1-daniel@iogearbox.net (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next,v2] wireguard: Wire-up big tcp support | expand |
Hi Daniel, On Fri, Oct 04, 2024 at 06:55:18PM +0200, Daniel Borkmann wrote: > Advertise GSO_MAX_SIZE as TSO max size in order support BIG TCP for wireguard. > This helps to improve wireguard performance a bit when enabled as it allows > wireguard to aggregate larger skbs in wg_packet_consume_data_done() via > napi_gro_receive(), but also allows the stack to build larger skbs on xmit > where the driver then segments them before encryption inside wg_xmit(). Thanks, I'll queue this up. Do you have any perf numbers on the speedup, btw? Jason
Hi Jason, On 10/7/24 6:42 PM, Jason A. Donenfeld wrote: > On Fri, Oct 04, 2024 at 06:55:18PM +0200, Daniel Borkmann wrote: >> Advertise GSO_MAX_SIZE as TSO max size in order support BIG TCP for wireguard. >> This helps to improve wireguard performance a bit when enabled as it allows >> wireguard to aggregate larger skbs in wg_packet_consume_data_done() via >> napi_gro_receive(), but also allows the stack to build larger skbs on xmit >> where the driver then segments them before encryption inside wg_xmit(). > > Thanks, I'll queue this up. Do you have any perf numbers on the speedup, > btw? Awesome, thanks! Few weeks back we had an LPC session with some initial work around wireguard performance (https://lpc.events/event/18/contributions/1968/) including benchmarks. It depends on the specific scenario, but this as well as an L2 GRO implementation (yet to come as patches in the next weeks) we got ~+15% for host-host in case of TCP STREAM. In case of Cilium I haven't done the measurement yet, but I expect that there would be further gains as bigger packets can traverse the stack and thus less processing in upper layers (e.g. our tcx BPF code). Thanks, Daniel
diff --git a/drivers/net/wireguard/device.c b/drivers/net/wireguard/device.c index 45e9b908dbfb..79be517b2216 100644 --- a/drivers/net/wireguard/device.c +++ b/drivers/net/wireguard/device.c @@ -301,6 +301,7 @@ static void wg_setup(struct net_device *dev) /* We need to keep the dst around in case of icmp replies. */ netif_keep_dst(dev); + netif_set_tso_max_size(dev, GSO_MAX_SIZE); memset(wg, 0, sizeof(*wg)); wg->dev = dev;
Advertise GSO_MAX_SIZE as TSO max size in order support BIG TCP for wireguard. This helps to improve wireguard performance a bit when enabled as it allows wireguard to aggregate larger skbs in wg_packet_consume_data_done() via napi_gro_receive(), but also allows the stack to build larger skbs on xmit where the driver then segments them before encryption inside wg_xmit(). Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Jason A. Donenfeld <Jason@zx2c4.com> Cc: Anton Protopopov <aspsk@isovalent.com> Cc: Martynas Pumputis <m@lambda.lt> --- v2: fixed up my gitconfig and Cc's now drivers/net/wireguard/device.c | 1 + 1 file changed, 1 insertion(+)