Hi,
> Right now qxl_render_update checks if the displaysurface buffer is not
> shared, meaning it was allocated by qemu, and in this case it replaces
> it with the flipped buffer.
I think we should first reqire spice-server 0.8.latest, so
update_area_complete is available unconditionally. Then do any
displaysurface updates in that callback (or a bh kicked by that
callback). Handle both shared & non-shared cases. I think we also can
get rid of the flip buffer then and just use a non-shared displaysurface
in that case (and flip the upside-down qxl surface while copying to the
qemu displaysurface).
> But right after that surface->data gets reset, by vga_hw_screen_dump:
>
> vga_hw_screen_dump/console_select/qemu_resize_displaysurface/ds->allocator->resize_displaysurface/defaultallocator_resize_displaysurface/qemu_alloc_display
>
> Hence my line of thought that replacing the allocator with my own would
> prevent this. Since you have misgivings about using our own allocator
> that I don't know how to resolve, I'm instead doing a second
> reallocation in our dpy_resize callback qxl.c:display_resize, in affect
> it means that we have three allocations and three deallocations for every
> screendump. Do you still think it's less ugly then an allocator? note
> that I have sdl and vnc working with spice with my allocator scheme.
> (just didn't test all three together yet).
IMHO that calls for a patch like this to get rid of the pointless
console_select() call:
--- a/console.c
+++ b/console.c
@@ -181,12 +181,14 @@ void vga_hw_screen_dump(const char *filename)
/* There is currently no way of specifying which screen we want to
dump,
so always dump the first one. */
- console_select(0);
+ if (previous_active_console && previous_active_console->index != 0) {
+ console_select(0);
+ }
if (consoles[0] && consoles[0]->hw_screen_dump) {
consoles[0]->hw_screen_dump(consoles[0]->hw, filename);
}
- if (previous_active_console) {
+ if (previous_active_console && previous_active_console->index != 0) {
console_select(previous_active_console->index);
}
}
cheers,
Gerd
