Re: Should linuxkpi linux_file_ioctl support flexible array copy?

On Sun, 15 Feb 2026, yi zishun wrote:

> Hi,
> Thanks for the questions!
>
>
> On Sat, Feb 14, 2026 at 11:20:58PM +0000, Bjoern A. Zeeb wrote:
>>
>> Let me ask two totally different questions:
>>
>> (a) is this udmabuf or u-dma-buf?
>
> It is the mainline linux udmabuf driver
> (linux/drivers/dma-buf/udmabuf.c)
>
>
>> (b) what are the expected consumers for your code?
>
> The primary consumer is QEMU. When passing the `-device
> virtio-gpu-pci,blob=true` parameter to launch QEMU, it allows the host
> compositor to import the DMA-BUF directly (zero-copy), instead of having
> to do an explicit GPU upload when using shared memory (shm).
>
> It is a project I see in the SummerOfCodeIdeas page. Details are in
> https://wiki.freebsd.org/SummerOfCodeIdeas#Port_or_reimplement_udmabuf

Ok, I'll leave it to Aymeric as I do not fully understand how all these things
are layered and glued together.

I was mostly curious to find out if/why this would be LinuxKPI code, or why
we would need the LinuxKPI interface/structures for this?

I am currently assuming that the main benefit would be if using some
virtio-drm driver and wanting to display the guest FB locally on the host
or in a GPU VF/PF setup rather than streaming it away?  But there's a
missing piece to the puzzle for me still.

That said, it all sounds totally exciting :)


-- 
Bjoern A. Zeeb                                                     r15:7