Re: [PATCH v2 1/6] hw/xen: Set XenBackendInstance in the XenDevice before realizing it

[Paul Durrant <xadimgnik@xxxxxxxxx>]

On 21/11/2023 22:10, Volodymyr Babchuk wrote:
> From: David Woodhouse <dwmw@xxxxxxxxxxxx>
>
> This allows a XenDevice implementation to know whether it was created
> by QEMU, or merely discovered in XenStore after the toolstack created
> it. This will allow us to create frontend/backend nodes only when we
> should, rather than unconditionally attempting to overwrite them from
> a driver domain which doesn't have privileges to do so.
>
> As an added benefit, it also means we no longer have to call the
> xen_backend_set_device() function from the device models immediately
> after calling qdev_realize_and_unref(). Even though we could make
> the argument that it's safe to do so, and the pointer to the unreffed
> device *will* actually still be valid, it still made my skin itch to
> look at it.
>
> Signed-off-by: David Woodhouse <dwmw@xxxxxxxxxxxx>
> ---
>   hw/block/xen-block.c         | 3 +--
>   hw/char/xen_console.c        | 2 +-
>   hw/net/xen_nic.c             | 2 +-
>   hw/xen/xen-bus.c             | 4 ++++
>   include/hw/xen/xen-backend.h | 2 --
>   include/hw/xen/xen-bus.h     | 2 ++
>   6 files changed, 9 insertions(+), 6 deletions(-)
Actually, I think you should probably update 
xen_backend_try_device_destroy() in this patch too. It currently uses 
xen_backend_list_find() to check whether the specified XenDevice has an 
associated XenBackendInstance.
  Paul