libkrunfw: Add virtio-media driver support

[dorindabassey]

Add virtio-media driver patches v4 from upstream kernel mailing list to enable vhost-user media device support in libkrun. The driver proxies V4L2 operations between guest userspace and the virtio host backend, supporting MMAP buffer sharing through a shared memory region.

Changes:

• Add patches 0031-0036 from [PATCH v8 ] media: Add virtio-media driver (lore.kernel.org/linux-media)
• Enable media kernel configs (CONFIG_MEDIA_SUPPORT, CONFIG_MEDIA_CAMERA_SUPPORT, CONFIG_VIDEO_DEV, CONFIG_MEDIA_VIRTIO)

Fixes (kernel 6.12 backport):

• v4l2_fh_add()/v4l2_fh_del() signature: drop second argument
  (mainline changed to 2-arg in 6.13+)
• kzalloc_obj() does not exist in 6.12: replace with
  kzalloc(sizeof(*session), GFP_KERNEL)
• file_to_v4l2_fh() does not exist in 6.12: add compat macro
  in virtio_media.h

The driver has been tested with rust-vmm vhost-device-media backend and a webcam, capturing video through the full SHMEM MMAP path.

Upstream patch authors:

Alexandre Courbot gnurou@gmail.com
Brian Daniels briandaniels@google.com

[aesteve-rh]

The VIRTIO_ID_MEDIA patch is available in https://lore.kernel.org/all/20260310-virtio-media-id-v1-1-be211bcf682b@redhat.com/

[dorindabassey]

Thanks Albert! I could pick it up and drop the local patch.

[slp]

@dorindabassey Any idea why is it failing on aarch64?

[dorindabassey]

@dorindabassey Any idea why is it failing on aarch64?

There were two separate issues:

1. virtio_media.h UAPI header bugs (all architectures):
   The v7 patches had UAPI header issues:

• Used u32/u64 instead of __u32/__u64
• Missing #include <linux/types.h>
• C++-style comments in header guards

These was failing make headers_check on all architectures. I've fixed them in the backport (patch 0032) and will send upstream review.

2. videodev2.h struct timespec issue (aarch64/riscv64 only):
   This one was not a virtio-media bug - it's a kernel 6.12 videodev2.h issue that only manifested on aarch64/riscv64. During headers_check, struct timespec was incomplete because <sys/time.h> doesn't provide it on those architectures. I've added a workaround patch (0031) using __kernel_old_timespec as fallback.

CI is passing now with both fixes applied.

[aesteve-rh]

Btw, it changes nothing, but the commit has made its way upstream: torvalds/linux@2187be2

In case you want to remove the patch here.

[dorindabassey]

about that patch, I'd still leave it here, we're on an older kernel, but we can drop the patch if we update the kernel.

[dorindabassey]

Updated to v8