Virtio-GPU Xres and Yres seettings

Virtio-GPU Xres and Yres seettings

[Pratik Parvati]

[-- Attachment #1: Type: text/plain, Size: 940 bytes --]

Hi Team,

I am working on a Virtio-GPU device (backend) for one of our customer
projects - we are using the Virtio-GPU driver (frontend) to drive our
device. Our device code has been written using Qemu virtio-gpu device code
as a reference. Our device is setting the resolution to 1024x768 as a
response to the command received from the driver (i.e,
VIRTIO_GPU_CMD_GET_DISPLAY_INFO) - I also verified that driver is receiving
the response bytes properly, but the next
command (i.e, VIRTIO_GPU_CMD_RESOURCE_CREATE_2D) that device receive has a
width and height as 4096 and 2160 respectively. The driver is assuming a 4k
display for some reason and I need your help to understand why this is
happening.

Our display is corrupted for the same Xres and Yres misunderstanding
between the device and the driver - Please guide me to understand the
problem. Let me know if you have any questions or need more information.

Thanks & Regards,
Pratik

[-- Attachment #2: Type: text/html, Size: 1799 bytes --]

Re: Virtio-GPU Xres and Yres seettings

[Philippe Mathieu-Daudé]

Cc'ing Gerd & Marc-André.

On 12/20/21 05:26, Pratik Parvati wrote:
> Hi Team,
> 
> I am working on a Virtio-GPU device (backend) for one of our customer
> projects - we are using the Virtio-GPU driver (frontend) to drive our
> device. Our device code has been written using Qemu virtio-gpu device
> code as a reference. Our device is setting the resolution to 1024x768 as
> a response to the command received from the driver (i.e,
> VIRTIO_GPU_CMD_GET_DISPLAY_INFO) - I also verified that driver is
> receiving the response bytes properly, but the next
> command (i.e, VIRTIO_GPU_CMD_RESOURCE_CREATE_2D) that device receive has
> a width and height as 4096 and 2160 respectively. The driver is assuming
> a 4k display for some reason and I need your help to understand why this
> is happening.
> 
> Our display is corrupted for the same Xres and Yres misunderstanding
> between the device and the driver - Please guide me to understand the
> problem. Let me know if you have any questions or need more information.
> 
> Thanks & Regards,
> Pratik

Re: Virtio-GPU Xres and Yres seettings

[Gerd Hoffmann]

On Mon, Dec 20, 2021 at 09:55:00AM +0100, Philippe Mathieu-Daudé wrote:
> Cc'ing Gerd & Marc-André.
> 
> On 12/20/21 05:26, Pratik Parvati wrote:
> > Hi Team,
> > 
> > I am working on a Virtio-GPU device (backend) for one of our customer
> > projects - we are using the Virtio-GPU driver (frontend) to drive our
> > device. Our device code has been written using Qemu virtio-gpu device
> > code as a reference. Our device is setting the resolution to 1024x768 as
> > a response to the command received from the driver (i.e,
> > VIRTIO_GPU_CMD_GET_DISPLAY_INFO) - I also verified that driver is
> > receiving the response bytes properly, but the next
> > command (i.e, VIRTIO_GPU_CMD_RESOURCE_CREATE_2D) that device receive has
> > a width and height as 4096 and 2160 respectively. The driver is assuming
> > a 4k display for some reason and I need your help to understand why this
> > is happening.

Wild guess: virtio_gpu_resp_display_info.pmodes[0].enabled is false?

> > Our display is corrupted for the same Xres and Yres misunderstanding
> > between the device and the driver

Oh, the VIRTIO_GPU_CMD_GET_DISPLAY_INFO reply is just a hint for the
driver.  The driver is free to choose whatever it wants, your device
must be prepared for that.

The linux kms driver specifically returns a long list of standard modes
(including 4k) to userspace, with the resolution returned by
VIRTIO_GPU_CMD_GET_DISPLAY_INFO tagged as 'preferred'.  Typically
userspace (aka display server, aka xorg / wayland / ...) uses the
preferred resolution by default, but userspace can also pick something
else and the kernel driver will forward the request to the device.

So, it might also be the software running inside the guest which
picks 4k ...

HTH & take care,
  Gerd

Re: Virtio-GPU Xres and Yres seettings

[Pratik Parvati]

[-- Attachment #1: Type: text/plain, Size: 4045 bytes --]

> > Cc'ing Gerd & Marc-André.
> >
> > On 12/20/21 05:26, Pratik Parvati wrote:
> > > Hi Team,
> > >
> > > I am working on a Virtio-GPU device (backend) for one of our customer
> > > projects - we are using the Virtio-GPU driver (frontend) to drive our
> > > device. Our device code has been written using Qemu virtio-gpu device
> > > code as a reference. Our device is setting the resolution to 1024x768
as
> > > a response to the command received from the driver (i.e,
> > > VIRTIO_GPU_CMD_GET_DISPLAY_INFO) - I also verified that driver is
> > > receiving the response bytes properly, but the next
> > > command (i.e, VIRTIO_GPU_CMD_RESOURCE_CREATE_2D) that device receive
has
> > > a width and height as 4096 and 2160 respectively. The driver is
assuming
> > > a 4k display for some reason and I need your help to understand why
this
> > > is happening.
>
> Wild guess: virtio_gpu_resp_display_info.pmodes[0].enabled is false?

virtio_gpu_resp_display_info.pmodes[0].enabled is true, I have verified
this.

> > > Our display is corrupted for the same Xres and Yres misunderstanding
> > > between the device and the driver
>
> Oh, the VIRTIO_GPU_CMD_GET_DISPLAY_INFO reply is just a hint for the
> driver.  The driver is free to choose whatever it wants, your device
> must be prepared for that.
>
> The linux kms driver specifically returns a long list of standard modes
> (including 4k) to userspace, with the resolution returned by
> VIRTIO_GPU_CMD_GET_DISPLAY_INFO tagged as 'preferred'.  Typically
> userspace (aka display server, aka xorg / wayland / ...) uses the
> preferred resolution by default, but userspace can also pick something
> else and the kernel driver will forward the request to the device.
>
> So, it might also be the software running inside the guest which
> picks 4k ...

Is it possible that EDID bytes are sent wrong to the driver (as a response
to VIRTIO_GPU_CMD_GET_EDID command)?? - I am configuring these bytes the
same way Qemu does.

Thanks for Quick response.

Regards,
Pratik
Regards,
Pratik


On Mon, 20 Dec 2021 at 16:06, Gerd Hoffmann <kraxel@redhat.com> wrote:

> On Mon, Dec 20, 2021 at 09:55:00AM +0100, Philippe Mathieu-Daudé wrote:
> > Cc'ing Gerd & Marc-André.
> >
> > On 12/20/21 05:26, Pratik Parvati wrote:
> > > Hi Team,
> > >
> > > I am working on a Virtio-GPU device (backend) for one of our customer
> > > projects - we are using the Virtio-GPU driver (frontend) to drive our
> > > device. Our device code has been written using Qemu virtio-gpu device
> > > code as a reference. Our device is setting the resolution to 1024x768
> as
> > > a response to the command received from the driver (i.e,
> > > VIRTIO_GPU_CMD_GET_DISPLAY_INFO) - I also verified that driver is
> > > receiving the response bytes properly, but the next
> > > command (i.e, VIRTIO_GPU_CMD_RESOURCE_CREATE_2D) that device receive
> has
> > > a width and height as 4096 and 2160 respectively. The driver is
> assuming
> > > a 4k display for some reason and I need your help to understand why
> this
> > > is happening.
>
> Wild guess: virtio_gpu_resp_display_info.pmodes[0].enabled is false?
>
> > > Our display is corrupted for the same Xres and Yres misunderstanding
> > > between the device and the driver
>
> Oh, the VIRTIO_GPU_CMD_GET_DISPLAY_INFO reply is just a hint for the
> driver.  The driver is free to choose whatever it wants, your device
> must be prepared for that.
>
> The linux kms driver specifically returns a long list of standard modes
> (including 4k) to userspace, with the resolution returned by
> VIRTIO_GPU_CMD_GET_DISPLAY_INFO tagged as 'preferred'.  Typically
> userspace (aka display server, aka xorg / wayland / ...) uses the
> preferred resolution by default, but userspace can also pick something
> else and the kernel driver will forward the request to the device.
>
> So, it might also be the software running inside the guest which
> picks 4k ...
>
> HTH & take care,
>   Gerd
>
>

[-- Attachment #2: Type: text/html, Size: 6282 bytes --]

Re: Virtio-GPU Xres and Yres seettings

[Gerd Hoffmann]

  Hi,

> Is it possible that EDID bytes are sent wrong to the driver (as a response
> to VIRTIO_GPU_CMD_GET_EDID command)?? - I am configuring these bytes the
> same way Qemu does.

EDID is optional, so you can try disable the EDID feature bit and see
what happens.

You can also check /sys/class/drm/card0-${output}/edid inside the
guest.

take care,
  Gerd

Re: Virtio-GPU Xres and Yres seettings

[Pratik Parvati]

[-- Attachment #1: Type: text/plain, Size: 1251 bytes --]

> EDID is optional, so you can try disable the EDID feature bit and see
> what happens.

Thanks Gerd, after disabling the EDID, I was able to get the required
resolution (basically width and height) from the driver.

Another strange observation - When the device receives the
command VIRTIO_GPU_CMD_RESOURCE_ATTACH_BACKING with the number of
entries having a pixel data in scatter gather format, the device is trying
to store these bytes in contiguous memory. When I read those sg memory, the
device receives all zeros from the driver (for a 1024x768 display,
the device receives 3MB of data from the driver). Is this an expected
behaviour? - If not, what is the driver trying to display on the screen?

This is happening during linux boot on the guest machine.

Thanks & Regards,
Pratik


On Mon, 20 Dec 2021 at 17:48, Gerd Hoffmann <kraxel@redhat.com> wrote:

>   Hi,
>
> > Is it possible that EDID bytes are sent wrong to the driver (as a
> response
> > to VIRTIO_GPU_CMD_GET_EDID command)?? - I am configuring these bytes the
> > same way Qemu does.
>
> EDID is optional, so you can try disable the EDID feature bit and see
> what happens.
>
> You can also check /sys/class/drm/card0-${output}/edid inside the
> guest.
>
> take care,
>   Gerd
>
>

[-- Attachment #2: Type: text/html, Size: 2567 bytes --]

Re: Virtio-GPU Xres and Yres seettings

[Gerd Hoffmann]

On Mon, Dec 20, 2021 at 10:44:06PM +0530, Pratik Parvati wrote:
> > EDID is optional, so you can try disable the EDID feature bit and see
> > what happens.
> 
> Thanks Gerd, after disabling the EDID, I was able to get the required
> resolution (basically width and height) from the driver.
> 
> Another strange observation - When the device receives the
> command VIRTIO_GPU_CMD_RESOURCE_ATTACH_BACKING with the number of
> entries having a pixel data in scatter gather format, the device is trying
> to store these bytes in contiguous memory. When I read those sg memory, the
> device receives all zeros from the driver (for a 1024x768 display,
> the device receives 3MB of data from the driver). Is this an expected
> behaviour? - If not, what is the driver trying to display on the screen?

How about reading the virtio spec?
display updates are handled with transfer and flush commands.

take care,
  Gerd

Re: Virtio-GPU Xres and Yres seettings

[Pratik Parvati]

[-- Attachment #1: Type: text/plain, Size: 1298 bytes --]

I apologise for not putting the question properly. I am referring virtio
spec to understand the driver and device operation - I had a few questions
on pixel data, as it was not working as expected.

My display is working fine now; thanks for your help.

Regards,
Pratik


On Tue, 21 Dec 2021 at 16:41, Gerd Hoffmann <kraxel@redhat.com> wrote:

> On Mon, Dec 20, 2021 at 10:44:06PM +0530, Pratik Parvati wrote:
> > > EDID is optional, so you can try disable the EDID feature bit and see
> > > what happens.
> >
> > Thanks Gerd, after disabling the EDID, I was able to get the required
> > resolution (basically width and height) from the driver.
> >
> > Another strange observation - When the device receives the
> > command VIRTIO_GPU_CMD_RESOURCE_ATTACH_BACKING with the number of
> > entries having a pixel data in scatter gather format, the device is
> trying
> > to store these bytes in contiguous memory. When I read those sg memory,
> the
> > device receives all zeros from the driver (for a 1024x768 display,
> > the device receives 3MB of data from the driver). Is this an expected
> > behaviour? - If not, what is the driver trying to display on the screen?
>
> How about reading the virtio spec?
> display updates are handled with transfer and flush commands.
>
> take care,
>   Gerd
>
>

[-- Attachment #2: Type: text/html, Size: 2253 bytes --]