Re: [REGRESSION] aquantia: Sunshine/Moonlight UDP video streaming broken since 5b4015ad833c ("net: aquantia: Remove redundant UDP length adjustment with GSO_PARTIAL")

[Matthew Schwartz]

> On May 5, 2026, at 8:05 AM, Gal Pressman <gal@xxxxxxxxxx> wrote:
> 
> On 27/04/2026 21:26, Matthew Schwartz wrote:
>>> On 4/27/26 11:09 AM, Gal Pressman wrote:
>>> Hello Matthew,
>>> 
>>> On 27/04/2026 2:20, Matthew Schwartz wrote:
>>>> Hello,
>>>> 
>>>> When using a previously working setup of remote streaming from my workstation to another device via Sunshine (the host server) and Moonlight (the client app) on my home network, I no longer receive any video output on the client app after upgrading my host workstation to kernel 7.0. Reverting back to kernel 6.19 on the host restored my setup to a working state.
>>>> 
>>>> After bisecting, I landed on 5b4015ad833c ("net: aquantia: Remove redundant UDP length adjustment with GSO_PARTIAL") as the first bad commit. I confirmed this by moving the cable to my second on-board NIC (Intel) on the same workstation, which restored video output without any other kernel changes. My affected on-board NIC is Aquantia AQC113 [1d6a:04c0] (rev 03), atlantic driver, firmware 1.3.34, MTU 1500.
>>>> 
>>>> Looking into it a bit further, ethtool -K enp97s0 tx-udp-segmentation off also serves as a workaround on my Aquantia port without changing to my other ethernet port. The working Intel NIC reports tx-udp-segmentation as "off [fixed]", so traffic falls back to software UDP segmentation on there.
>>>> 
>>>> Please let me know if there's any additional info I can provide.
>>>> 
>>>> Thanks,
>>>> Matt
>>>> 
>>>> #regzbot introduced: 5b4015ad833c
>>> 
>>> Thank you for the report and the bisect!
>>> 
>>> I will take a look and try to figure out what's wrong (though I don't
>>> have real hardware to test on).
>>> Is the userspace app open source? can I see its code and try to run it
>>> myself?
>> 
>> Thanks for the reply. The code for Sunshine is available here: https://github.com/LizardByte/Sunshine and the code for Moonlight is here: https://github.com/moonlight-stream/moonlight-qt.
>> 
>> I have been using the Arch Linux Sunshine package which I installed by following the Linux instructions here: https://docs.lizardbyte.dev/projects/sunshine/latest/md_docs_2getting__started.html, but there are also binaries for other distros or it's buildable from source. For Moonlight, I have been using the Flatpak distributed on Flathub because the client device runs an atomic rootfs, but you can also use any other device that Moonlight supports.
>> 
>>> 
>>> I will be OOO for the rest of the week, hope to have some meaningful
>>> reply by the end of next week.
>> 
> 
> I think I see the issue, do you mind testing the following diff?

Sorry for the delay. I applied this diff and I’m able to stream from my host to clients again without any issues.

If you send this version upstream, feel free to add my Tested-by to it.

Thanks,
Matt

> 
> index a0813d425b71..5bd1706b11b0 100644
> --- a/net/ipv4/udp_offload.c
> +++ b/net/ipv4/udp_offload.c
> @@ -599,10 +599,22 @@ struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
>        uh = udp_hdr(seg);
>    }
> 
> -    /* last packet can be partial gso_size, account for that in checksum */
> -    newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
> -               seg->data_len);
> -    check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
> +    if (skb_is_gso(seg)) {
> +        newlen = msslen;
> +    } else {
> +        /* last packet can be partial gso_size, account for that in
> +         * checksum.
> +         */
> +        newlen = htons(skb_tail_pointer(seg) -
> +                   skb_transport_header(seg) + seg->data_len);
> +        check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
> +    }
> 
>    uh->len = newlen;
>    uh->check = check;