[PATCH] virtio-gpu: create drm fd based on specified render node path

[PATCH] virtio-gpu: create drm fd based on specified render node path

[Luqmaan Irshad]

Added a special callback function called virtio_get_drm_fd to create
a render node based on the path specified by the user via QEMU command
line.  This function is called during the virglrenderer callback sequence
where we specify the get_drm_fd function pointer to call back our
new function, allowing us to pass the fd of our created render node.

Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com

Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
---
 hw/display/virtio-gpu-gl.c     |  4 ++++
 hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
 include/hw/virtio/virtio-gpu.h |  1 +
 3 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/hw/display/virtio-gpu-gl.c b/hw/display/virtio-gpu-gl.c
index b640900fc6..c0587929e4 100644
--- a/hw/display/virtio-gpu-gl.c
+++ b/hw/display/virtio-gpu-gl.c
@@ -161,6 +161,10 @@ static const Property virtio_gpu_gl_properties[] = {
                     VIRTIO_GPU_FLAG_VENUS_ENABLED, false),
     DEFINE_PROP_BIT("drm_native_context", VirtIOGPU, parent_obj.conf.flags,
                     VIRTIO_GPU_FLAG_DRM_ENABLED, false),
+#if VIRGL_RENDERER_CALLBACKS_VERSION >= 2s
+    DEFINE_PROP_STRING("drm_render_node", VirtIOGPU,
+                       parent_obj.drm_render_node),
+#endif
 };
 
 static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
diff --git a/hw/display/virtio-gpu-virgl.c b/hw/display/virtio-gpu-virgl.c
index 6ad3b46066..c271194260 100644
--- a/hw/display/virtio-gpu-virgl.c
+++ b/hw/display/virtio-gpu-virgl.c
@@ -1016,6 +1016,15 @@ void virtio_gpu_virgl_process_cmd(VirtIOGPU *g,
     }
 }
 
+#if VIRGL_RENDERER_CALLBACKS_VERSION >= 2
+static int virtio_get_drm_fd(void *opaque)
+{
+    VirtIOGPU *g = opaque;
+    int fd = open(g->parent_obj.drm_render_node, O_RDWR);
+    return fd;
+}
+#endif
+
 static void virgl_write_fence(void *opaque, uint32_t fence)
 {
     VirtIOGPU *g = opaque;
@@ -1300,7 +1309,13 @@ int virtio_gpu_virgl_init(VirtIOGPU *g)
     }
     if (virtio_gpu_drm_enabled(g->parent_obj.conf)) {
         flags |= VIRGL_RENDERER_DRM;
-
+#if VIRGL_RENDERER_CALLBACKS_VERSION >= 2
+        if (g->parent_obj.drm_render_node) {
+            if (virtio_gpu_3d_cbs.version < 2)
+                virtio_gpu_3d_cbs.version = 2;
+            virtio_gpu_3d_cbs.get_drm_fd = virtio_get_drm_fd;
+        }
+#endif
         if (!(flags & VIRGL_RENDERER_ASYNC_FENCE_CB)) {
             /*
              * Virglrenderer skips enabling DRM context support without
diff --git a/include/hw/virtio/virtio-gpu.h b/include/hw/virtio/virtio-gpu.h
index a4963508a4..aba8343848 100644
--- a/include/hw/virtio/virtio-gpu.h
+++ b/include/hw/virtio/virtio-gpu.h
@@ -152,6 +152,7 @@ struct VirtIOGPUBase {
     struct virtio_gpu_base_conf conf;
     struct virtio_gpu_config virtio_config;
     const GraphicHwOps *hw_ops;
+    char *drm_render_node;
 
     int renderer_blocked;
     int enable;
-- 
2.43.0

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Dmitry Osipenko]

Hi,

On 12/5/25 21:49, Luqmaan Irshad wrote:
> Added a special callback function called virtio_get_drm_fd to create
> a render node based on the path specified by the user via QEMU command
> line.  This function is called during the virglrenderer callback sequence
> where we specify the get_drm_fd function pointer to call back our
> new function, allowing us to pass the fd of our created render node.
> 
> Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com
> 
> Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
> ---
>  hw/display/virtio-gpu-gl.c     |  4 ++++
>  hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
>  include/hw/virtio/virtio-gpu.h |  1 +
>  3 files changed, 21 insertions(+), 1 deletion(-)

Do you think it could be possible and worthwhile to make QEMU's EGL
display to use same GPU as virgl automatically? I.e. we tell QEMU/EGL
which GPU to use and then virgl will use same DRM device that backs EGL.

-- 
Best regards,
Dmitry

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Irshad, Luqmaan]

[-- Attachment #1: Type: text/plain, Size: 2211 bytes --]

[AMD Official Use Only - AMD Internal Distribution Only]

Hello Dmitry,

Thats an interesting idea!  I would be curious to look into that further, however my ability to do so may be limited since my contract ends at the end of this week.

After my contract ends, I won't have access to a multi-GPU system I can test this on.  I may be able to work out some way to borrow the system, but at the moment this is uncertain.

If this patch looks good, my suggestion would be to merge this change, and we can look into adding this as a separate patch on top of this if I can get access to a system.  Let me know how this sounds 🙂

Sincerely,

Luq
________________________________
From: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Sent: Sunday, December 7, 2025 7:49 PM
To: Irshad, Luqmaan <Luqmaan.Irshad@amd.com>; qemu-devel@nongnu.org <qemu-devel@nongnu.org>
Cc: mst@redhat.com <mst@redhat.com>; alex.bennee@linaro.org <alex.bennee@linaro.org>; odaki@rsg.ci.i.u-tokyo.ac.jp <odaki@rsg.ci.i.u-tokyo.ac.jp>; Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>; luckmaan21@gmail.com <luckmaan21@gmail.com>
Subject: Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

Hi,

On 12/5/25 21:49, Luqmaan Irshad wrote:
> Added a special callback function called virtio_get_drm_fd to create
> a render node based on the path specified by the user via QEMU command
> line.  This function is called during the virglrenderer callback sequence
> where we specify the get_drm_fd function pointer to call back our
> new function, allowing us to pass the fd of our created render node.
>
> Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com
>
> Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
> ---
>  hw/display/virtio-gpu-gl.c     |  4 ++++
>  hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
>  include/hw/virtio/virtio-gpu.h |  1 +
>  3 files changed, 21 insertions(+), 1 deletion(-)

Do you think it could be possible and worthwhile to make QEMU's EGL
display to use same GPU as virgl automatically? I.e. we tell QEMU/EGL
which GPU to use and then virgl will use same DRM device that backs EGL.

--
Best regards,
Dmitry

[-- Attachment #2: Type: text/html, Size: 5209 bytes --]

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Dmitry Osipenko]

On 12/9/25 21:57, Irshad, Luqmaan wrote:
> [AMD Official Use Only - AMD Internal Distribution Only]
> 
> Hello Dmitry,
> 
> Thats an interesting idea!  I would be curious to look into that further, however my ability to do so may be limited since my contract ends at the end of this week.
> 
> After my contract ends, I won't have access to a multi-GPU system I can test this on.  I may be able to work out some way to borrow the system, but at the moment this is uncertain.
> 
> If this patch looks good, my suggestion would be to merge this change, and we can look into adding this as a separate patch on top of this if I can get access to a system.  Let me know how this sounds 🙂
> 
> Sincerely,

Thanks for notifying that you won't work on the patch further. It will
be merged or amended.

-- 
Best regards,
Dmitry

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Akihiko Odaki]

On 2025/12/08 9:49, Dmitry Osipenko wrote:
> Hi,
> 
> On 12/5/25 21:49, Luqmaan Irshad wrote:
>> Added a special callback function called virtio_get_drm_fd to create
>> a render node based on the path specified by the user via QEMU command
>> line.  This function is called during the virglrenderer callback sequence
>> where we specify the get_drm_fd function pointer to call back our
>> new function, allowing us to pass the fd of our created render node.

I guess what you need can be achieved by specifying a render node for 
the display. Headless displays (egl-headless and dbus) has the 
rendernode property for this. For the other displays, Mesa should choose 
an appropriate render node, and it can be overridden with the DRI_PRIME 
environment variable:
https://docs.mesa3d.org/envvars.html#envvar-DRI_PRIME

>>
>> Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com
>>
>> Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
>> ---
>>   hw/display/virtio-gpu-gl.c     |  4 ++++
>>   hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
>>   include/hw/virtio/virtio-gpu.h |  1 +
>>   3 files changed, 21 insertions(+), 1 deletion(-)
> 
> Do you think it could be possible and worthwhile to make QEMU's EGL
> display to use same GPU as virgl automatically? I.e. we tell QEMU/EGL
> which GPU to use and then virgl will use same DRM device that backs EGL.

As far as I understand, it is already ensured that virgl uses the EGL 
display QEMU uses, and I think that what you want. Opening a different 
render node and passing the node to virglrenderer breaks it.

Regards,
Akihiko Odaki

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Irshad, Luqmaan]

[-- Attachment #1: Type: text/plain, Size: 2729 bytes --]

[AMD Official Use Only - AMD Internal Distribution Only]

Hello Akihiko,

I just had a brief conversation with Pierre-Eric, and I believe the use case is slightly different than the one described.  For virgl, selecting using DRI_PRIME only would suffice because QEMU and VirGL (both using OpenGL), both rely on the driver to pickup the GPU.

For Native context however (this case) there is no GL driver use from virglrenderer side, which is why we inform virglrenderer using the callback functions instead.

Sincerely,

Luq

________________________________
From: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
Sent: Thursday, December 11, 2025 9:19:22 a.m.
To: Dmitry Osipenko <dmitry.osipenko@collabora.com>; Irshad, Luqmaan <Luqmaan.Irshad@amd.com>; qemu-devel@nongnu.org <qemu-devel@nongnu.org>
Cc: mst@redhat.com <mst@redhat.com>; alex.bennee@linaro.org <alex.bennee@linaro.org>; Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>; luckmaan21@gmail.com <luckmaan21@gmail.com>
Subject: Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

On 2025/12/08 9:49, Dmitry Osipenko wrote:
> Hi,
>
> On 12/5/25 21:49, Luqmaan Irshad wrote:
>> Added a special callback function called virtio_get_drm_fd to create
>> a render node based on the path specified by the user via QEMU command
>> line.  This function is called during the virglrenderer callback sequence
>> where we specify the get_drm_fd function pointer to call back our
>> new function, allowing us to pass the fd of our created render node.

I guess what you need can be achieved by specifying a render node for
the display. Headless displays (egl-headless and dbus) has the
rendernode property for this. For the other displays, Mesa should choose
an appropriate render node, and it can be overridden with the DRI_PRIME
environment variable:
https://docs.mesa3d.org/envvars.html#envvar-DRI_PRIME

>>
>> Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com
>>
>> Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
>> ---
>>   hw/display/virtio-gpu-gl.c     |  4 ++++
>>   hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
>>   include/hw/virtio/virtio-gpu.h |  1 +
>>   3 files changed, 21 insertions(+), 1 deletion(-)
>
> Do you think it could be possible and worthwhile to make QEMU's EGL
> display to use same GPU as virgl automatically? I.e. we tell QEMU/EGL
> which GPU to use and then virgl will use same DRM device that backs EGL.

As far as I understand, it is already ensured that virgl uses the EGL
display QEMU uses, and I think that what you want. Opening a different
render node and passing the node to virglrenderer breaks it.

Regards,
Akihiko Odaki


[-- Attachment #2: Type: text/html, Size: 5387 bytes --]

Re: [PATCH] virtio-gpu: create drm fd based on specified render node path

[Akihiko Odaki]

On 2025/12/12 1:30, Irshad, Luqmaan wrote:
> [AMD Official Use Only - AMD Internal Distribution Only]

Your keep sending message with this "Internal Distribution Only" mark. 
Can you remove it? Perhaps you may ask Honglei Huang, who is also from 
AMD and work on virtio-gpu/QEMU.

> 
> Hello Akihiko,
> 
> I just had a brief conversation with Pierre-Eric, and I believe the use case is slightly different than the one described.  For virgl, selecting using DRI_PRIME only would suffice because QEMU and VirGL (both using OpenGL), both rely on the driver to pickup the GPU.

DRI_PRIME is used for EGL, not OpenGL. EGL is agnostic to rendering APIs 
so it works for Vulkan too.

> 
> For Native context however (this case) there is no GL driver use from virglrenderer side, which is why we inform virglrenderer using the callback functions instead.

Now I understand the problem you are facing; virglrenderer does not 
respect the EGL display for DRM native context, which is the problem 
that needs to be fixed; adding another callback function only works 
around it.

Fortunately, there is EXT_device_drm_render_node, an EGL extension that 
allows virglrenderer to query the underlying render node from the EGL 
display, which can be used for DRM native context:
https://registry.khronos.org/EGL/extensions/EXT/EGL_EXT_device_drm_render_node.txt

Consuming the EGL display, you can exploit the render-API-agnostic 
nature of EGL and ensure DRM native context will work consistently with 
VirGL and Venus. virglrenderer will be able to use the appropriate 
render node chosen by Mesa, and if you want to choose a render node by 
your own, you will be able to use the rendernode property of headless 
displays or DRI_PRIME for the other displays; no need to set another 
property for virtio-gpu.

Using EXT_device_drm_render_node on the virglrenderer side may also 
benefit other VMMs that use virglrenderer.

Regards,
Akihiko Odaki

> 
> Sincerely,
> 
> Luq
> 
> ________________________________
> From: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
> Sent: Thursday, December 11, 2025 9:19:22 a.m.
> To: Dmitry Osipenko <dmitry.osipenko@collabora.com>; Irshad, Luqmaan <Luqmaan.Irshad@amd.com>; qemu-devel@nongnu.org <qemu-devel@nongnu.org>
> Cc: mst@redhat.com <mst@redhat.com>; alex.bennee@linaro.org <alex.bennee@linaro.org>; Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>; luckmaan21@gmail.com <luckmaan21@gmail.com>
> Subject: Re: [PATCH] virtio-gpu: create drm fd based on specified render node path
> 
> On 2025/12/08 9:49, Dmitry Osipenko wrote:
>> Hi,
>>
>> On 12/5/25 21:49, Luqmaan Irshad wrote:
>>> Added a special callback function called virtio_get_drm_fd to create
>>> a render node based on the path specified by the user via QEMU command
>>> line.  This function is called during the virglrenderer callback sequence
>>> where we specify the get_drm_fd function pointer to call back our
>>> new function, allowing us to pass the fd of our created render node.
> 
> I guess what you need can be achieved by specifying a render node for
> the display. Headless displays (egl-headless and dbus) has the
> rendernode property for this. For the other displays, Mesa should choose
> an appropriate render node, and it can be overridden with the DRI_PRIME
> environment variable:
> https://docs.mesa3d.org/envvars.html#envvar-DRI_PRIME
> 
>>>
>>> Based-on: 20251116125641.2255794-1-dmitry.osipenko@collabora.com
>>>
>>> Signed-off-by: Luqmaan Irshad <luirshad@amd.com>
>>> ---
>>>    hw/display/virtio-gpu-gl.c     |  4 ++++
>>>    hw/display/virtio-gpu-virgl.c  | 17 ++++++++++++++++-
>>>    include/hw/virtio/virtio-gpu.h |  1 +
>>>    3 files changed, 21 insertions(+), 1 deletion(-)
>>
>> Do you think it could be possible and worthwhile to make QEMU's EGL
>> display to use same GPU as virgl automatically? I.e. we tell QEMU/EGL
>> which GPU to use and then virgl will use same DRM device that backs EGL.
> 
> As far as I understand, it is already ensured that virgl uses the EGL
> display QEMU uses, and I think that what you want. Opening a different
> render node and passing the node to virglrenderer breaks it.
> 
> Regards,
> Akihiko Odaki
> 
>