Benno Fünfstück <[email protected]> writes:
> Hi Marc-André,
>
> On Mon, 13 Apr 2026 at 15:23, Marc-André Lureau <[email protected]>
> wrote:
>
> Arguably, this is a bug in virgl vrend_renderer_get_cursor_content(),
> since the function doesn't return y0_top it should always return the
> data in the correct orientation.
>
> We may have a temporary workaround in qemu though for older virgl, but
> I think it should be solved there too.
>
> Thanks, I didn't know about this. I submitted a MR to virglrenderer now as
> well to fix it there:
> https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/1611
I see this is merged now. Do we want to version gate the fix or just
drop it and wait for things to be backported?
>
> How should I change the patch to make it a temporary fix? Some kind of
> version gate (but not sure yet in which version of
> virglrenderer the fix will end up being release)?
>
>
> > ---
> > hw/display/virtio-gpu-gl.c | 20 +++++++++++++++++++-
> > 1 file changed, 19 insertions(+), 1 deletion(-)
> >
> > diff --git a/hw/display/virtio-gpu-gl.c b/hw/display/virtio-gpu-gl.c
> > index 2b7a41c466..4562d3405e 100644
> > --- a/hw/display/virtio-gpu-gl.c
> > +++ b/hw/display/virtio-gpu-gl.c
> > @@ -33,6 +33,9 @@ static void virtio_gpu_gl_update_cursor_data(VirtIOGPU
> *g,
> > VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
> > uint32_t width, height;
> > uint32_t pixels, *data;
> > + bool y_0_top = true;
> > + int ret;
> > + struct virgl_renderer_resource_info info;
> >
> > if (gl->renderer_state != RS_INITED) {
> > return;
> > @@ -49,8 +52,23 @@ static void virtio_gpu_gl_update_cursor_data(VirtIOGPU
> *g,
> > return;
> > }
> >
> > + memset(&info, 0, sizeof(info));
> > + ret = virgl_renderer_resource_get_info(resource_id, &info);
> > + if (ret == 0) {
> > + y_0_top = info.flags & VIRTIO_GPU_RESOURCE_FLAG_Y_0_TOP;
> > + }
> > +
> > pixels = s->current_cursor->width * s->current_cursor->height;
> > - memcpy(s->current_cursor->data, data, pixels * sizeof(uint32_t));
> > + if (y_0_top) {
> > + memcpy(s->current_cursor->data, data, pixels * sizeof(uint32_t));
> > + } else {
> > + uint32_t *dst = s->current_cursor->data;
> > + for (uint32_t y = 0; y < height; y++) {
> > + memcpy(dst + y * width,
> > + data + (height - 1 - y) * width,
> > + width * sizeof(uint32_t));
> > + }
> > + }
> > free(data);
> > }
> >
> > --
> > 2.50.1 (Apple Git-155)
> >
> >
> > --
> > Neodyme AG
> > Sitz der Gesellschaft / Address: Dirnismaning 55 | Halle 13 |
> > 85748 Garching b.München
> > Postanschrift: Rosenthaler Straße 72a | 10119
> > Berlin
> >
> > Registergericht / Registry court: München, HRB 269168
> > Vorstand /
> > Management Board: Thomas Lambertz | Tobias Madl
> > Aufsichtsratsvorsitzender /
> > Chairman of the Supervisory Board: Hendrik Hofstadt
> >
> >
> >
> >
> >
>
> --
> Marc-André Lureau
>
> Neodyme AG
>
> Sitz der Gesellschaft / Address: Dirnismaning 55 | Halle 13 | 85748 Garching
> b.München
> Postanschrift: Rosenthaler Straße 72a | 10119 Berlin
> Registergericht / Registry court: München, HRB 269168
> Vorstand / Management Board: Thomas Lambertz | Tobias Madl
> Aufsichtsratsvorsitzender / Chairman of the Supervisory Board: Hendrik
> Hofstadt
--
Alex Bennée
Virtualisation Tech Lead @ Linaro
