It _can_ be done. The small root process would be _part_ of the XF86
distribution - it would do all the mode switching for the X server. As
such it can do _anything_ the current X server does (which obviously
includes switching back to text mode).
XFree86 already had support for setting the video card clock speed with an
external program - this just takes it one step further, to set the whole
mode up.
> EXAMPLE
>
> shazam GoodCard xserver-GOOD -bpp 16 &
No. Example:
startx
and the X server does all of this. I didn't imply that the small program
would be _separate_ from the X server in any way: it would be a separate
process for security reasons, but it would be part of X.
> Works well for GoodCard and VeryGoodCard2000.
> Fails miserable on any card that can be set into
> a state such that to leave that state you must
> know what state it is in. For such cards the
> knowledge that it was once in a particular state
> is not useful.
>
> The cards for which shazam is not useful include:
> S3-yada-yada
> ATI-blah-blah
> some others
I'm going to ignore the rest of this thread, because I get responses like
this from people who haven't thought the problem through.
OF COURSE the small program has to keep track of the video mode. That is a
given. What's so hard about that?
For example, when the user presses "ctrl-alt-+" to get to another video
mode, the "real" X server would just send a signal back to its parent
telling the parent to switch to the next higher resolution. The X server
proper would never need to worry about the thing.
What's so hard to understand? The fact is that NONE of this actually
requires any kernel help.
The things that might require kernel help are things like DMA and
interrupt access, and there it makes perfect sense. I want to re-iterate
that I'm not against having the kernel help the X server as required.
What I AM against is these stupid people that think that the "kill -9 X"
argument is worth anything. It is not - because it is easily fixed by
having a separate part that cleans up after the X server. Go back and read
my mail.
Anybody who thinks that it is easier to do things like mode switching in
kernel mode is very seriously mistaken. Kernel programming is a LOT harder
than programming a trusted deamon, and it is a lot easier to get the
kernel part wrong. And when the kernel part is wrong, the end result is
something much worse than just a graphical screen that you can't get to do
anything.
Some people claim that the kernel part of GGI is very small compared to X,
and thus easy to prove right. So what? If it is so easy to prove right it
_still_ should be done in user mode if at all possible. And I have just
told you exactly _how_ it is possible.
So go, and sin no more.
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu