Re: [PATCH net-next 6/7] net: bcmgenet: pad transmit frames out of the packet ready window

[Nicolai Buchwitz]

On 5.10.2026 17:04, netdev-bot+sashiko@xxxxxxxxxx wrote:

> [...]

> At this commit, on kernels with 16K or larger pages and an MTU of
> 3809..3820, the threshold drops to 3584. Frames of 3585..3647 bytes 
> then
> get padded.
>
> Later in the series, "net: bcmgenet: reassemble jumbo frames from 
> status
> block fragments" raises max_mtu to ENET_MAX_JUMBO_MTU on all page 
> sizes.
> After that, any MTU of 3890 or more pads every frame of 3841..3903
> bytes. MTUs of 3809..3889 pad frames of 3585..3647 bytes. Nothing later
> in the series appears to fix this.
>
> Should frames that carry a trailer be kept out of this window some 
> other
> way, rather than by appending bytes?

I do not see a way. Any frame length can fall in the window, so moving 
the
threshold only moves the window. Keeping every frame below the threshold
caps the MTU at 3808 and removes the point of the series.

Worth noting the window is unreachable below MTU 3809. At the default 
MTU
the longest frame is 1532 against a threshold of 3840, so nothing is 
ever
padded. This only affects jumbo configurations!

The alternatives are dropping those frames or leaving the transmitter
stalled until the interface is reopened. Padding seemed the least bad, 
but
I will note the trailer limitation in the commit message.

> [...]

For the other findings I will respin anyway, so:

---
pw-bot: cr