On Fri, Sep 11, 2026 at 10:01:01AM -0700, Jakub Kicinski wrote:
> On Thu, 10 Sep 2026 22:06:59 -0700 Narcisa Vasile wrote:
> > Add four new test cases that validate coalescing
> > under increased size limits (BIG TCP):
>
> Looks like SW is struggling to aggregate enough. Do we need longer
> timeouts perhaps? Note that the debug kernel build includes
> kernel/configs/debug.config
>
> AI says:
>
> Thanks for adding the BIG TCP coalescing coverage to gro_sw.py. CI is
> flagging the new subtests as failing on our debug-kernel executor
> (vmksft-net-drv-dbg, gro-sw-py):
>
> not ok gro_sw.test.ipv4_big_tcp_data_same
> not ok gro_sw.test.ipv4_big_tcp_data_lrg_sml
> not ok gro_sw.test.ipv4_big_tcp_tcp_seq
> not ok gro_sw.test.ipv4_big_tcp_large_max
> (and the ipv6 equivalents)
>
> The rx side never fully coalesces the injected segments before the
> background receiver gives up and the test's retry loop (6 attempts) is
> exhausted, e.g.:
>
> Expected {68952 }, Total 1 packets
> Received {20280 [!=68952]40560 [!=0]8112 [!=0]}, Total 3 packets.
I've run the tests in a loop on a slower system running a debug kernel using
10x the previous flush timeout and I still noticed retries every 10-20
iterations, so I've settled on 15x higher timeout to account for variability
across different systems.
1..4
ok 1 gro_sw.test.ipv4_big_tcp_data_same
ok 2 gro_sw.test.ipv4_big_tcp_data_lrg_sml
ok 3 gro_sw.test.ipv4_big_tcp_tcp_seq
ok 4 gro_sw.test.ipv4_big_tcp_large_max
# Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0
1..4
ok 1 gro_sw.test.ipv6_big_tcp_data_same
ok 2 gro_sw.test.ipv6_big_tcp_data_lrg_sml
ok 3 gro_sw.test.ipv6_big_tcp_tcp_seq
ok 4 gro_sw.test.ipv6_big_tcp_large_max
# Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0
>
