Hi On Sat, Jan 20, 2024 at 4:54 AM Vivek Kasireddy <vivek.kasire...@intel.com> wrote: > > Since most encoders (invoked by Spice) may not work with tiled memory > associated with a texture, we need to create another texture that has > linear memory layout and use that instead. > > Note that, there does not seem to be a direct way to indicate to the > GL implementation that a texture's backing memory needs to be linear. > Instead, we have to do it in a roundabout way where we first need to > create a tiled texture and obtain a dmabuf fd associated with it via > EGL. Next, we have to create a memory object by importing the dmabuf > fd created earlier and finally create a linear texture by using the > memory object as the texture storage mechanism. > > Cc: Gerd Hoffmann <kra...@redhat.com> > Cc: Marc-André Lureau <marcandre.lur...@redhat.com> > Cc: Frediano Ziglio <fredd...@gmail.com> > Cc: Dongwon Kim <dongwon....@intel.com> > Signed-off-by: Vivek Kasireddy <vivek.kasire...@intel.com> > --- > ui/spice-display.c | 33 +++++++++++++++++++++++++++++++++ > 1 file changed, 33 insertions(+) > > diff --git a/ui/spice-display.c b/ui/spice-display.c > index 08b4aec921..94cb378dbe 100644 > --- a/ui/spice-display.c > +++ b/ui/spice-display.c > @@ -893,6 +893,7 @@ static void spice_gl_switch(DisplayChangeListener *dcl, > { > SimpleSpiceDisplay *ssd = container_of(dcl, SimpleSpiceDisplay, dcl); > EGLint stride, fourcc; > + GLuint texture = 0; > int fd; > > if (ssd->ds) { > @@ -912,6 +913,38 @@ static void spice_gl_switch(DisplayChangeListener *dcl, > return; > } > > + if (remote_client && surface_format(ssd->ds) != PIXMAN_r5g6b5) {
hmm > + /* > + * We really want to ensure that the memory layout of the texture > + * is linear; otherwise, the encoder's output may show > corruption. > + */ > + surface_gl_create_texture_from_fd(ssd->ds, fd, &texture); What if the encoder actually supports tiled layout? Shouldn't this conversion be done at the encoder level as necessary? It's also strange to reuse an FD associated with a tiled texture for a linear layout (I am uncomfortable with all this tbh) > + > + /* > + * A successful return after glImportMemoryFdEXT() means that > + * the ownership of fd has been passed to GL. In other words, > + * the fd we got above should not be used anymore. > + */ > + if (texture > 0) { > + fd = egl_get_fd_for_texture(texture, > + &stride, &fourcc, > + NULL); > + if (fd < 0) { I suggest adding warnings or tracing, to help debug issues... > + glDeleteTextures(1, &texture); > + fd = egl_get_fd_for_texture(ssd->ds->texture, > + &stride, &fourcc, > + NULL); > + if (fd < 0) { > + surface_gl_destroy_texture(ssd->gls, ssd->ds); > + return; > + } > + } else { > + surface_gl_destroy_texture(ssd->gls, ssd->ds); > + ssd->ds->texture = texture; Have you tried this series with virgl? (I doubt the renderer accepts that the scanout texture is changed) > + } > + } > + } > + > trace_qemu_spice_gl_surface(ssd->qxl.id, > surface_width(ssd->ds), > surface_height(ssd->ds), > -- > 2.39.2 > > -- Marc-André Lureau