Re: [RFC PATCH] media: Virtual camera driver

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Sun, Feb 01, 2026 at 10:54:34PM +0200, Jarkko Sakkinen wrote:
> On Sun, Feb 01, 2026 at 10:35:12PM +0200, Jarkko Sakkinen wrote:
> > On Sun, Feb 01, 2026 at 10:06:49PM +0200, Laurent Pinchart wrote:
> > > On Sun, Feb 01, 2026 at 09:04:00PM +0200, Jarkko Sakkinen wrote:
> > > > On Sun, Feb 01, 2026 at 08:20:11PM +0200, Laurent Pinchart wrote:
> > > > > On Sun, Feb 01, 2026 at 03:33:38PM +0200, Jarkko Sakkinen wrote:
> > > > > > vcam is a DMA-BUF backed virtual camera driver capable of creating video
> > > > > > capture devices to which data can be streamed through /dev/vcam after
> > > > > > calling VCAM_IOC_CREATE. Frames are pushed with VCAM_IOC_QUEUE and recycled
> > > > > > with VCAM_IOC_DEQUEUE.
> > > > > > 
> > > > > > Zero-copy semantics are supported for shared DMA-BUF between capture and
> > > > > > output.
> > > > > >
> > > > > > Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
> > > > > > ---
> > > > > > Early feedback e.g., is this completely in wrong direction? V4L2 world
> > > > > > is relatively alien world, and thus I need a sanity check ;-)
> > > > > 
> > > > > We already have multiple virtual drivers, including vivid and vimc.
> > > > > Could you please explain the rationale for yet another one, and why the
> > > > > new features it provides (if any) can't be added to existing drivers ?
> > > > 
> > > > There is a notable user base for v4l2-loopback. It is the defacto choice
> > > > for streaming phone cams.
> > > 
> > > This will then likely face the same hurdles as v4l2-loopback, the main
> > > one being that camera support should be upstreamed with proper drivers
> > > instead of a closed-source userspace daemon.
> > > 
> > > For phone cameras, the way forward upstream is libcamera. Until kernel
> > > drivers for ISPs are available, the soft ISP is a stop-gap solution. It
> > > recently gained GPU acceleration support (with work to improve image
> > > quality with additional algorithms ongoing).
> > 
> > That might have some weight as a pro but the unarguable con is that at
> > the same time this policy retains a base of tainted kernels in the wild.
> > 
> > Not saying that this weight more but it is important to remark this
> > fact.
> 
> It's widely packaged for different distributions and even embedded build
> systems forming across the board tained ecosystem. And this has been
> ongoing for years. Suggesting PipeWire as "a fix" for all possible
> situations is not "a solution".

PipeWire may not solve all of the world's problems, but it's part of a
clean solution for this particular issue. The fact that everybody relied
on an out-of-tree kernel module instead of designing a better stack is
not a reason to merge v4l2loopback upstream now that we have a better
option that is actively developed.

-- 
Regards,

Laurent Pinchart