Re: commit 662dc80a5e86 breaks rmnet over usb

[David Laight]

On Mon, 23 Feb 2026 15:08:41 +0100
Laurent Vivier <lvivier@xxxxxxxxxx> wrote:

...
> It's weird to be able to set an MTU that is not supported by the hardware.
> 
> To restore performance I think the rx_usb_size should be decoupled from MTU max in qmi_wwan.

A lot of 'usbnet' drivers support usb packets that contain multiple ethernet frames.
IIRC (I looked at this code quite a few years ago) some (all?) lie very badly
about the skb 'true_size' of receive packets - putting in the frame size
even though it is sat in 16kb+ of kernel memory.

In reality the driver want to receive the 'stream' of usb packets and
debatch it directly into ethernet frames without going though the 'urb'
buffers at all.
ISTR that would be possible with the xhci/usb3 hardware, but would be a
massive re-write of parts of the stack.
I was only trying (and I mean failing) to get 100M working reliably.
Project got canned for other reasons.

	David