> -----Original Message-----
> From: Neal Cardwell <[email protected]> 
> Sent: Tuesday, January 20, 2026 8:35 PM
> To: Jakub Kicinski <[email protected]>
> Cc: Chia-Yu Chang (Nokia) <[email protected]>; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; [email protected]; 
> [email protected]; [email protected]; Koen De Schepper (Nokia) 
> <[email protected]>; [email protected]; 
> [email protected]; [email protected]; cheshire 
> <[email protected]>; [email protected]; [email protected]; Vidhi Goel 
> <[email protected]>
> Subject: Re: [PATCH v9 net-next 15/15] selftests/net: packetdrill: add TCP 
> Accurate ECN cases
> 
> 
> CAUTION: This is an external email. Please be very careful when clicking 
> links or opening attachments. See the URL nok.it/ext for additional 
> information.
> 
> 
> 
> On Tue, Jan 20, 2026 at 1:53 PM Jakub Kicinski <[email protected]> wrote:
> >
> > On Mon, 19 Jan 2026 19:58:52 +0100 [email protected]
> > wrote:
> > > Linux Accurate ECN test sets using ACE counters and AccECN options 
> > > to cover several scenarios: Connection teardown, different ACK 
> > > conditions, counter wrapping, SACK space grabbing, fallback schemes, 
> > > negotiation retransmission/reorder/loss, AccECN option drop/loss, 
> > > different handshake reflectors, data with marking, and different sysctl 
> > > values.
> >
> > Thank you for closing the packetdrill side, and big thanks to Neal for 
> > prioritizing getting it reviewed and merged!
> >
> > I updated the packetdrill build in netdev CI and looks like one of the 
> > cases is flaking a little. Since it looks like you'll have to respin, 
> > please try to fix:
> >
> > # 1..2
> > # tcp_accecn_client_accecn_options_lost.pkt:32: error handling packet: 
> > timing error: expected outbound packet in relative time range 
> > +0.020000~+0.500000 sec but happened at +0.015816 sec # script packet:  
> > 0.181936 .5 1:1013(1012) ack 1 <ECN e1b 1 ceb 0 e0b 1,nop> # actual 
> > packet:  0.177752 .EA 1:1013(1012) ack 1 win 1050 <ECN e1b 1 ceb 0 e0b 
> > 1,nop> # not ok 1 ipv4 # tcp_accecn_client_accecn_options_lost.pkt:32: 
> > error handling packet: timing error: expected outbound packet in 
> > relative time range +0.020000~+0.500000 sec but happened at +0.015800 
> > sec # script packet:  0.181952 .5 1:1013(1012) ack 1 <ECN e1b 1 ceb 0 
> > e0b 1,nop> # actual packet:  0.177752 .EA 1:1013(1012) ack 1 win 1050 
> > <ECN e1b 1 ceb 0 e0b 1,nop> # not ok 2 ipv6 # # Totals: pass:0 fail:2 
> > xfail:0 xpass:0 skip:0 error:0
> >
> > https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fnetd
> > ev-ctrl.bots.linux.dev%2Flogs%2Fvmksft%2Fpacketdrill%2Fresults%2F48220
> > 1%2F115-tcp-accecn-client-accecn-options-lost-pkt%2Fstdout&data=05%7C0
> > 2%7Cchia-yu.chang%40nokia-bell-labs.com%7Cf125ccafc7134b620bad08de585b
> > 1a35%7C5d4717519675428d917b70f44f9630b0%7C0%7C0%7C639045345459758258%7
> > CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlA
> > iOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=kn3Pb
> > VXw%2Bkznnf7VaYzwP%2FL3IO2LYGYQzZOWS3HRZ6w%3D&reserved=0
> 
> Probably this is happening because the SRTT is around 56ms:
> 
> .050 * 7/8 + 1/8 * .1 = .05625 sec
> 
> So the RACK fast recovery starts afte rabout 15ms due to .25 * srtt being 
> about 14ms:
> (.050 * 7/8 + 1/8 * .1) * .25 = .0140625 sec
> 
> If we make the SRTT 100ms then the fast retransmit should be around:
> 
> (.1 * 7/8 + 1/8 * .1) * .25 = .025 sec
> 
> So I'd suggest changing the timing of the SYNACK from 50ms to 100ms:
> 
> old:
> +0.05 < [ect0] SW. 0:0(0) ack 1 win 32767 <mss 1024,ECN e0b 1 ceb 0
> e1b 1,nop,nop,nop,sackOK,nop,wscale 8>
> 
> new:
> +.1 < [ect0] SW. 0:0(0) ack 1 win 32767 <mss 1024,ECN e0b 1 ceb 0 e1b
> 1,nop,nop,nop,sackOK,nop,wscale 8>
> 
> neal

Thanks Neal and Eric, I've fixed this issue as well as the concerned raised in 
patch 10 of listen socket.
All AccECN packetdrill still pass from my end, so I will submit v10.

Thanks.
Chai-Yu

Reply via email to