From: Marc-André Lureau <marcandre.lureau@redhat.com>

dpy_get_ui_info() shouldn't be called if the underlying GPU doesn't
support it.

Before the assert() was added and the regression introduced, GTK code
used to get "zero" UI info, for ex with a simple VGA device. The assert
was added to prevent from calling when there are no console too. The
other display backend that calls dpy_get_ui_info() correctly checks that
pre-condition.

Calling dpy_set_ui_info() is "safe" in this case, it will simply return
an error that can be generally ignored.

Fixes: commit a92e7bb4c ("ui: add precondition for dpy_get_ui_info()")
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
---
 ui/gtk.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/ui/gtk.c b/ui/gtk.c
index e09f97a86b..7b542da0c0 100644
--- a/ui/gtk.c
+++ b/ui/gtk.c
@@ -726,6 +726,10 @@ static void gd_set_ui_refresh_rate(VirtualConsole *vc, int refresh_rate)
 {
     QemuUIInfo info;
 
+    if (!dpy_ui_info_supported(vc->gfx.dcl.con)) {
+        return;
+    }
+
     info = *dpy_get_ui_info(vc->gfx.dcl.con);
     info.refresh_rate = refresh_rate;
     dpy_set_ui_info(vc->gfx.dcl.con, &info, true);
@@ -735,6 +739,10 @@ static void gd_set_ui_size(VirtualConsole *vc, gint width, gint height)
 {
     QemuUIInfo info;
 
+    if (!dpy_ui_info_supported(vc->gfx.dcl.con)) {
+        return;
+    }
+
     info = *dpy_get_ui_info(vc->gfx.dcl.con);
     info.width = width;
     info.height = height;
-- 
2.41.0

15.09.2023 14:36, marcandre.lureau@redhat.com:
> From: Marc-André Lureau <marcandre.lureau@redhat.com>
> 
> dpy_get_ui_info() shouldn't be called if the underlying GPU doesn't
> support it.
> 
> Before the assert() was added and the regression introduced, GTK code
> used to get "zero" UI info, for ex with a simple VGA device. The assert
> was added to prevent from calling when there are no console too. The
> other display backend that calls dpy_get_ui_info() correctly checks that
> pre-condition.
> 
> Calling dpy_set_ui_info() is "safe" in this case, it will simply return
> an error that can be generally ignored.
> 
> Fixes: commit a92e7bb4c ("ui: add precondition for dpy_get_ui_info()")
> Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
> ---
>   ui/gtk.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/ui/gtk.c b/ui/gtk.c
> index e09f97a86b..7b542da0c0 100644
> --- a/ui/gtk.c
> +++ b/ui/gtk.c
> @@ -726,6 +726,10 @@ static void gd_set_ui_refresh_rate(VirtualConsole *vc, int refresh_rate)
>   {
>       QemuUIInfo info;
>   
> +    if (!dpy_ui_info_supported(vc->gfx.dcl.con)) {
> +        return;
> +    }
> +
>       info = *dpy_get_ui_info(vc->gfx.dcl.con);

Current dpy_ui_info_supported():

bool dpy_ui_info_supported(QemuConsole *con)
{
     if (con == NULL) {
         con = active_console;
     }
     if (con == NULL) {
         return false;
     }

     return con->hw_ops->ui_info != NULL;
}

This whole thing smells a bit wrong.  I'm not saying it *is* wrong, but
the feeling is here.

Where dpy_ui_info_supported() is called with a NULL con, so that an "active"
console needs to be used instead?

Maybe we should instead use something like

    if (!vc->gfx.dcl.con)

here in gd_set_ui_refresh_rate() ?

At the very least, the code is a bit, well, confusing.

/mjt

Hi

On Fri, Sep 15, 2023 at 4:46 PM Michael Tokarev <mjt@tls.msk.ru> wrote:
>
> 15.09.2023 14:36, marcandre.lureau@redhat.com:
> > From: Marc-André Lureau <marcandre.lureau@redhat.com>
> >
> > dpy_get_ui_info() shouldn't be called if the underlying GPU doesn't
> > support it.
> >
> > Before the assert() was added and the regression introduced, GTK code
> > used to get "zero" UI info, for ex with a simple VGA device. The assert
> > was added to prevent from calling when there are no console too. The
> > other display backend that calls dpy_get_ui_info() correctly checks that
> > pre-condition.
> >
> > Calling dpy_set_ui_info() is "safe" in this case, it will simply return
> > an error that can be generally ignored.
> >
> > Fixes: commit a92e7bb4c ("ui: add precondition for dpy_get_ui_info()")
> > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
> > ---
> >   ui/gtk.c | 8 ++++++++
> >   1 file changed, 8 insertions(+)
> >
> > diff --git a/ui/gtk.c b/ui/gtk.c
> > index e09f97a86b..7b542da0c0 100644
> > --- a/ui/gtk.c
> > +++ b/ui/gtk.c
> > @@ -726,6 +726,10 @@ static void gd_set_ui_refresh_rate(VirtualConsole *vc, int refresh_rate)
> >   {
> >       QemuUIInfo info;
> >
> > +    if (!dpy_ui_info_supported(vc->gfx.dcl.con)) {
> > +        return;
> > +    }
> > +
> >       info = *dpy_get_ui_info(vc->gfx.dcl.con);
>
> Current dpy_ui_info_supported():
>
> bool dpy_ui_info_supported(QemuConsole *con)
> {
>      if (con == NULL) {
>          con = active_console;
>      }
>      if (con == NULL) {
>          return false;
>      }
>
>      return con->hw_ops->ui_info != NULL;
> }
>
> This whole thing smells a bit wrong.  I'm not saying it *is* wrong, but
> the feeling is here.
>
> Where dpy_ui_info_supported() is called with a NULL con, so that an "active"
> console needs to be used instead?
>
> Maybe we should instead use something like
>
>     if (!vc->gfx.dcl.con)
>
> here in gd_set_ui_refresh_rate() ?
>
> At the very least, the code is a bit, well, confusing.

I agree it's a bit confusing to have dpy_*() functions accept NULL
arguments to mean "active_console". I don't really have a good idea
what to do else. Just having "if (!vc->gfx.dcl.con)" is not quite
enough though, as we shouldn't get_ui_info() if it's not supported
(there are no errors returned then, just "invalid/zero" values).

-- 
Marc-André Lureau