[2/2] ui: introduce vfio_display_reset

Submitted by Gerd Hoffmann on April 23, 2018, 1:12 p.m.

Details

Message ID 20180423131202.zcyam25lkh7khbil@sirius.home.kraxel.org
State New
Headers show
Series "Add vfio display reset handling" ( rev: 2 ) in Intel GVT devel

Not browsing as part of any series.

Commit Message

Gerd Hoffmann April 23, 2018, 1:12 p.m.
> +    surface = qemu_console_surface(vdev->dpy->con);

> +    dpy_gl_scanout_disable(vdev->dpy->con);
> +    dpy_gfx_switch_surface(vdev->dpy->con, surface);

Hmm, so you ask which surface is active, then switch to it.
Why?

Maybe you just need a full display update after disabling the gl
scanout?

cheers,
  Gerd

PS: maybe adding a dpy_gfx_update_full() helper is useful:

Patch hide | download patch | download mbox

diff --git a/include/ui/console.h b/include/ui/console.h
index 37a8d68d29..981b519dde 100644
--- a/include/ui/console.h
+++ b/include/ui/console.h
@@ -291,6 +291,7 @@  bool dpy_ui_info_supported(QemuConsole *con);
 int dpy_set_ui_info(QemuConsole *con, QemuUIInfo *info);
 
 void dpy_gfx_update(QemuConsole *con, int x, int y, int w, int h);
+void dpy_gfx_update_full(QemuConsole *con);
 void dpy_gfx_replace_surface(QemuConsole *con,
                              DisplaySurface *surface);
 void dpy_text_cursor(QemuConsole *con, int x, int y);
diff --git a/ui/console.c b/ui/console.c
index 3fb2f4e09f..b02510cdca 100644
--- a/ui/console.c
+++ b/ui/console.c
@@ -1574,6 +1574,16 @@  void dpy_gfx_update(QemuConsole *con, int x, int y, int w, int h)
     }
 }
 
+void dpy_gfx_update_full(QemuConsole *con)
+{
+    if (!con->surface) {
+        return;
+    }
+    dpy_gfx_update(con, 0, 0,
+                   surface_width(con->surface),
+                   surface_height(con->surface));
+}
+
 void dpy_gfx_replace_surface(QemuConsole *con,
                              DisplaySurface *surface)
 {

Comments

> -----Original Message-----

> From: intel-gvt-dev [mailto:intel-gvt-dev-bounces@lists.freedesktop.org] On

> Behalf Of Gerd Hoffmann

> Sent: Monday, April 23, 2018 9:12 PM

> To: Zhang, Tina <tina.zhang@intel.com>

> Cc: alex.williamson@redhat.com; intel-gvt-dev@lists.freedesktop.org; qemu-

> devel@nongnu.org; zhenyuw@linux.intel.com; Yuan, Hang

> <hang.yuan@intel.com>

> Subject: Re: [PATCH 2/2] ui: introduce vfio_display_reset

> 

> > +    surface = qemu_console_surface(vdev->dpy->con);

> 

> > +    dpy_gl_scanout_disable(vdev->dpy->con);

> > +    dpy_gfx_switch_surface(vdev->dpy->con, surface);

> 

> Hmm, so you ask which surface is active, then switch to it.

> Why?


What this patch-set is going to do is to disable scanout mode during guest OS reset, so that Qemu UI (e.g. gtk) can directly use QemuConsole's DisplaySurface instead of guest dma-buf framebuffer during the guest OS reset process. 
gd_egl_draw is invodked when the Qemu UI window gets the focus. If this process is happened during guest OS reboot, gd_egl_scanout_flush() will be invoked (as vc->gfx.scanout_mode==true), which uses guest dma-buf framebuffer. See the following logic in gtk-egl.c:
void gd_egl_draw(VirtualConsole *vc)
{
    GdkWindow *window;
    int ww, wh;

    if (!vc->gfx.gls) {
        return;
    }

    if (vc->gfx.scanout_mode) {
        gd_egl_scanout_flush(&vc->gfx.dcl, 0, 0, vc->gfx.w, vc->gfx.h);
    } else {
        if (!vc->gfx.ds) {
            return;
        }
        eglMakeCurrent(qemu_egl_display, vc->gfx.esurface,
                       vc->gfx.esurface, vc->gfx.ectx);

        window = gtk_widget_get_window(vc->gfx.drawing_area);
        gdk_drawable_get_size(window, &ww, &wh);
        surface_gl_setup_viewport(vc->gfx.gls, vc->gfx.ds, ww, wh);
        surface_gl_render_texture(vc->gfx.gls, vc->gfx.ds);

        eglSwapBuffers(qemu_egl_display, vc->gfx.esurface);
    }
}

Using guest dma-buf framebuffer during guest OS reboot will lead to host GPU hang, as the GGTT is invalid during guest OS reboot.
That's why we want to disable the scanout mode and release all the dma-buf resources during guest OS reset.

After reviewing this patch-set again, I think we might not need the proposed dpy_gfx_switch_surface() any more.
The reason I proposed it was because I thought gfx.ds is changed to the surface related to guest dma-buf framebuffer and we need to switch it with QemuConsole's DisplaySurgface during reset.
But after reviewing the code, vc->gfx.ds is always pointing to the valid QemuConsoles' DisplaySurface. So we don’t need dpy_gfx_switch_surface() to take care of it.
Thanks.

> 

> Maybe you just need a full display update after disabling the gl scanout?

Do you still think updating full display is needed? It seems the existing vfio_display_dmabuf_update can be helpful enough to create a new texture if needed.
Thanks.


BR,
Tina
> 

> cheers,

>   Gerd

> 

> PS: maybe adding a dpy_gfx_update_full() helper is useful:

> 

> diff --git a/include/ui/console.h b/include/ui/console.h index

> 37a8d68d29..981b519dde 100644

> --- a/include/ui/console.h

> +++ b/include/ui/console.h

> @@ -291,6 +291,7 @@ bool dpy_ui_info_supported(QemuConsole *con);  int

> dpy_set_ui_info(QemuConsole *con, QemuUIInfo *info);

> 

>  void dpy_gfx_update(QemuConsole *con, int x, int y, int w, int h);

> +void dpy_gfx_update_full(QemuConsole *con);

>  void dpy_gfx_replace_surface(QemuConsole *con,

>                               DisplaySurface *surface);  void dpy_text_cursor(QemuConsole

> *con, int x, int y); diff --git a/ui/console.c b/ui/console.c index

> 3fb2f4e09f..b02510cdca 100644

> --- a/ui/console.c

> +++ b/ui/console.c

> @@ -1574,6 +1574,16 @@ void dpy_gfx_update(QemuConsole *con, int x, int

> y, int w, int h)

>      }

>  }

> 

> +void dpy_gfx_update_full(QemuConsole *con) {

> +    if (!con->surface) {

> +        return;

> +    }

> +    dpy_gfx_update(con, 0, 0,

> +                   surface_width(con->surface),

> +                   surface_height(con->surface)); }

> +

>  void dpy_gfx_replace_surface(QemuConsole *con,

>                               DisplaySurface *surface)  {

> 

> _______________________________________________

> intel-gvt-dev mailing list

> intel-gvt-dev@lists.freedesktop.org

> https://lists.freedesktop.org/mailman/listinfo/intel-gvt-dev
Hi,

> After reviewing this patch-set again, I think we might not need the
> proposed dpy_gfx_switch_surface() any more.  The reason I proposed it
> was because I thought gfx.ds is changed to the surface related to
> guest dma-buf framebuffer and we need to switch it with QemuConsole's
> DisplaySurgface during reset.  But after reviewing the code,
> vc->gfx.ds is always pointing to the valid QemuConsoles'
> DisplaySurface. So we don’t need dpy_gfx_switch_surface() to take care
> of it.

Good, this is what I expected.

> > Maybe you just need a full display update after disabling the gl
> > scanout?
> Do you still think updating full display is needed?

It certainly isn't required, gtk wouldn't access the dma-buf any more
after scanout_disable().

Without display update the last guest display may remain visible until
the guest reloads the driver though.  With display update the gtk ui
should show the (blank) DisplaySurface instead.  So it is more
cosmetical.

cheers,
  Gerd
> -----Original Message-----

> From: intel-gvt-dev [mailto:intel-gvt-dev-bounces@lists.freedesktop.org] On

> Behalf Of Gerd Hoffmann

> Sent: Tuesday, April 24, 2018 4:16 PM

> To: Zhang, Tina <tina.zhang@intel.com>

> Cc: alex.williamson@redhat.com; intel-gvt-dev@lists.freedesktop.org; qemu-

> devel@nongnu.org; zhenyuw@linux.intel.com; Yuan, Hang

> <hang.yuan@intel.com>

> Subject: Re: [PATCH 2/2] ui: introduce vfio_display_reset

> 

>   Hi,

> 

> > After reviewing this patch-set again, I think we might not need the

> > proposed dpy_gfx_switch_surface() any more.  The reason I proposed it

> > was because I thought gfx.ds is changed to the surface related to

> > guest dma-buf framebuffer and we need to switch it with QemuConsole's

> > DisplaySurgface during reset.  But after reviewing the code,

> > vc->gfx.ds is always pointing to the valid QemuConsoles'

> > DisplaySurface. So we don’t need dpy_gfx_switch_surface() to take care

> > of it.

> 

> Good, this is what I expected.

> 

> > > Maybe you just need a full display update after disabling the gl

> > > scanout?

> > Do you still think updating full display is needed?

> 

> It certainly isn't required, gtk wouldn't access the dma-buf any more after

> scanout_disable().

> 

> Without display update the last guest display may remain visible until the guest

> reloads the driver though.  With display update the gtk ui should show the

> (blank) DisplaySurface instead.  So it is more cosmetical.

Indeed. Thanks for the comments. I will send out the next version of this patch-set.
Thanks.

BR,
Tina
> 

> cheers,

>   Gerd

> 

> _______________________________________________

> intel-gvt-dev mailing list

> intel-gvt-dev@lists.freedesktop.org

> https://lists.freedesktop.org/mailman/listinfo/intel-gvt-dev