[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5e98e9666bfeb275ec168df24bb8e9a33781229e.camel@cyberus-technology.de>
Date: Wed, 08 Jan 2020 18:08:28 +0200
From: Julian Stecklina <julian.stecklina@...erus-technology.de>
To: Jani Nikula <jani.nikula@...ux.intel.com>,
intel-gvt-dev@...ts.freedesktop.org
Cc: linux-kernel@...r.kernel.org, hang.yuan@...el.com,
dri-devel@...ts.freedesktop.org, zhiyuan.lv@...el.com
Subject: Re: [PATCH 2/3] drm/i915/gvt: make gvt oblivious of kvmgt data
structures
On Wed, 2020-01-08 at 12:24 +0200, Jani Nikula wrote:
> On Mon, 06 Jan 2020, Julian Stecklina <julian.stecklina@...erus-technology.de>
> wrote:
[...]
> > + /* Hypervisor-specific device state. */
> > + void *vdev;
>
> I have no clue about the relative merits of the patch, but you can use
> the actual type for the pointer with a forward declaration. You don't
> need the definition for that.
>
> i.e.
>
> struct kvmgt_vdev;
> ...
> struct kvmgt_vdev *vdev;
The goal here is to make the GVT code independent of the hypervisor backend.
Different hypervisor backends need to keep different per-device state, so using
the KVM type here defeats the purpose.
I assume this is not only useful for us, but also for other hypervisor backends,
such as Xen or 3rd-party hypervisors.
Julian
Powered by blists - more mailing lists