Re: [PATCH v2] hw/display/tcx: Mark the VRAM dirty upon reset

[Philippe Mathieu-Daudé]

On Sun, Feb 6, 2022 at 11:44 AM Peter Maydell <peter.maydell@linaro.org> wrote:
>
> On Sun, 6 Feb 2022 at 09:30, Mark Cave-Ayland
> <mark.cave-ayland@ilande.co.uk> wrote:
> >
> > On 05/02/2022 15:39, Peter Maydell wrote:
> > > The handling of the vram buffer seems weird in this device overall,
> > > though -- the memory block is divided into three parts
> > >   * main vram, one byte per pixel
> > >   * vram24, four bytes per pixel
> > >   * cplane, four bytes per pixel
> > >
> > > As far as I can see, only if depth=24 (fixed at device creation
> > > time) do we use the vram24 and cplane parts. But we create the
> > > memory block at the same size regardless of depth and we expose
> > > the vram24 and cplane parts to the guest as sysbus MMIO regions
> > > that are mapped into guest memory regardless of depth...
> >
> > (goes and looks)
> >
> > It does look a bit odd certainly. Without Blue Swirl being around all I can 
> > only
> > guess as to why everything is configured to use an alias onto a single VRAM 
> > memory
> > region :/
> >
> > As for exposing the vram24 and cplane parts, I don't think it matters since 
> > 24-bit
> > mode is clearly designed to be backwards compatible with 8-bit mode. During
> > initialisation OpenBIOS reads the colour depth using the fw_cfg interface 
> > and adds
> > the registers for that mode into the DT as required so the correct 
> > information is
> > exposed to the guest.
>
> A guest won't notice if we expose stuff to it that it doesn't expect
> to be there -- but if the 8-bit-only device is not supposed to have
> those memory regions we shouldn't be creating them...

I guess I'll go with your code suggestion as a first step, and we can figure
out whether this MR is pointless and remove it later.