Re: [Qemu-devel] [PATCH] display: stop using DT_NOGRAPHIC, use DT_NONE

[Michael Tokarev]

28.06.2013 15:55, Peter Maydell wrote:
> On 28 June 2013 12:50, Michael Tokarev <address@hidden> wrote:

>> This ctrl+c handling is only about when you explicitly redirected
>> some other char device to guest, such as serial port.  Which don't
>> have much to do with display I think, hence I don't understand the
>> logic here.
> 
> The point is that one of the things -nographic does, as well
> as disabling the graphics display, is to redirect the serial
> port (among other things) to stdio. That's why there's code
> in the char backend that cares about "did you say -nographic":
> because -nographic means "no graphic display, serial port on
> terminal, send ctrl-c in terminal to guest rather than killing
> qemu, [some other stuff]".

That's exactly why I don't think looking at -nographic here in
serial code is wrong.  It should do the same regardless of
-nographic - if, say, serial is redirected to stdio, it should
always pass Ctrl+C to guest instead of killing it.  So I'm
not really sure we should test for -display none here, either.

> It would be nice to be able to say "-nographic is equivalent to
> '-display none -serial stdio -disable-ctrl-c -other-things'"
> but I'm not sure exactly what the -other-things are or even
> if we can set them all on the command line specifically.

We have some code in sun4 which also enabled serial ports
differently with and without -nographic, maybe because of
the same thing (which is what this all is about!).

And there's also another place, passing this NOGRAPHIC thing
to bios.

So far that's all.  With chardev you can explicitly control
Ctrl+C behavour using signal={on|off} attribute.


28.06.2013 15:56, Andreas Färber wrote:
[]
> The logic is that a) -nographic does have serial I/O on stdio and b)
> emulated serial ports may get repurposed (change order/usage) for this
> to work. CC'ing Alex and Blue who might remember more details.
>
> So in the end it boils down to whether we consider stdio a "display" and
> whether we may want to split this out into its own global variable to
> cleanly separate it from graphical DT_* options.

This is still wrong because it works differently if we explicitly
redirected serial port to stdio and used -display none, or it was
due to -nographic.



Thanks,

/mjt