Re: [RFC PATCH net-next v2] net: gro: coalesce short IPv4 packets padded to the minimum frame size

On Fri, Aug 14, 2026 at 12:26 AM Glenn Judd <[email protected]> wrote:
>
> Software GRO fails to coalesce a small IPv4 segment that was
> padded up to the 60-byte minimum Ethernet frame.
>
> The selftest tools/testing/selftests/drivers/net/gro.py subtest
> sw_ipv4_data_lrg_1byte sends {100, 1} expecting to receive {101}.
> In current code, it receives {100, 1} (no coalescing) instead.
>
> Cause: inet_gro_receive() computes its flush term from
> tot_len ^ skb_gro_len() before skb_gro_pull(), while skb_gro_len()
> still includes trailing Ethernet padding. A small IPv4 segment
> padded up to the 60-byte minimum frame has tot_len != skb_gro_len(),
> so flush is set and the runt never coalesces.
>
> v1 detected the padding with an added iph->tot_len read and
> skb_gro_len() comparison on every IPv4 GRO packet. Instead, split
> inet_gro_receive() so that everything after the header validation
> takes the flush term as a parameter, and pass a literal 0 on the
> common path. That folds away both flush updates and lets the
> transport dispatch become a tail call, leaving the common path
> shorter than before this patch rather than merely unchanged.
>
> Assisted-by: Claude:claude-opus-5
> Assisted-by: Codex:gpt-5.6
> Assisted-by: Meta:internal-AI-tooling
> Signed-off-by: Glenn Judd <[email protected]>
> ---

I am pretty sure I met NIC where the extra bytes were not zero.
Unfortunately I no longer have access to them.

Your patch seems to target a specific NIC.

Look at my prior commit : I made it generic. Checking if the bytes are
zero has the same cost as computing their partial checksum.
(real cost is the cache line miss)

commit 88078d98d1bb085d72af8437707279e203524fa5
Author: Eric Dumazet <[email protected]>
Date:   Wed Apr 18 11:43:15 2018 -0700

    net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends

    After working on IP defragmentation lately, I found that some large
    packets defeat CHECKSUM_COMPLETE optimization because of NIC adding
    zero paddings on the last (small) fragment.

    While removing the padding with pskb_trim_rcsum(), we set skb->ip_summed
    to CHECKSUM_NONE, forcing a full csum validation, even if all prior
    fragments had CHECKSUM_COMPLETE set.

    We can instead compute the checksum of the part we are trimming,
    usually smaller than the part we keep.

    Signed-off-by: Eric Dumazet <[email protected]>
    Signed-off-by: David S. Miller <[email protected]>