Re: [PATCH net-next] selftests: txtimestamp: ignore the old skb from ERRQUEUE

[Jakub Kicinski <kuba@kernel.org>]

On Mon,  3 Mar 2025 16:04:04 +0800 Jason Xing wrote:
> When I was playing with txtimestamp.c to see how kernel behaves,
> I saw the following error outputs if I adjusted the loopback mtu to
> 1500 and then ran './txtimestamp -4 -L 127.0.0.1 -l 30000 -t 100000':
> 
> test SND
>     USR: 1740877371 s 488712 us (seq=0, len=0)
>     SND: 1740877371 s 489519 us (seq=29999, len=1106)  (USR +806 us)
>     USR: 1740877371 s 581251 us (seq=0, len=0)
>     SND: 1740877371 s 581970 us (seq=59999, len=8346)  (USR +719 us)
>     USR: 1740877371 s 673855 us (seq=0, len=0)
>     SND: 1740877371 s 674651 us (seq=89999, len=30000)  (USR +795 us)
>     USR: 1740877371 s 765715 us (seq=0, len=0)
> ERROR: key 89999, expected 119999
> ERROR: -12665 us expected between 0 and 100000
>     SND: 1740877371 s 753050 us (seq=89999, len=1106)  (USR +-12665 us)
>     SND: 1740877371 s 800783 us (seq=119999, len=30000)  (USR +35068 us)
>     USR-SND: count=5, avg=4945 us, min=-12665 us, max=35068 us
> 
> Actually, the kernel behaved correctly after I did the analysis. The
> second skb carrying 1106 payload was generated due to tail loss probe,
> leading to the wrong estimation of tskey in this C program.
> 
> This patch does:
> - Neglect the old tskey skb received from ERRQUEUE.
> - Add a new test to count how many valid skbs received to compare with
> cfg_num_pkts.

This appears to break some UDP test cases when running in the CI:

https://netdev-3.bots.linux.dev/vmksft-net/results/16721/41-txtimestamp-sh/stdout
-- 
pw-bot: cr