> I as a user can do "kill -9 xserver" and my console is dead. This is not
then, don't do this! X has an exit handler to restore console in case
of crashes which works pretty good (but of course could be improved).
and if you still like shooting into your foot, why not start X again
in /etc/inittab ?? there are so many possibilities to deal with
problems and ideas in UNIX that putting stuff into kernel (which should
be kept small and simple if possible!) isn't always the first choice.
> good. Surely what we want is mode swapping and graphics card state
> information in the kernel. We can then link a new SAK function for use in
> emergencies, which can be _guarateed_ to get the console back to a useable
> state.
>
> It gets even nicer than that; if X dies horrible, its fd to (Eg.)
> /dev/graph gets closed, we the kernel notice this and restore console mode
> accordingly.
>
> Does anyone disagree about that or think it isn't needed???? If so I am
> _very_ interested to hear.
while it's a nice feature and I'd be happy to see it, it's not highly _needed_ IMHO.
I have worked on XFree86 setup for S3 cards for a verylong time now
and I know pretty well why I don't want this code in kernel if not
needed for _very_ good reasons.
> People are dithering and arguing about performance; one vital point seems
> to have been missed, and that is that a gfx card is supposed to deliver
> performance by performing acceleration operations while the CPU gets on
> with something more useful. Well under X currently it wastes my very fast
> CPU by polling the damn card to see when its bitBLT request is done. A
> nice queue system in the kernel can be envisaged where the kernel monitors
> the gfx card interrupts and automagically loads the next request from the
> queue asyncronously to the normal processor. The X-server can then sleep
> waiting for some "queue purged" message from the kernel.... etc.. you get
> the point.
sounds nice but you still have to prove that overhead of interrupts
and kernel calls isn't in the same order of magnitute (or much larger)
than user mode X server with polling (if needed at all). until you
show big performace boosts, that's all pure theory...
Harald
--
All SCSI disks will from now on ___ _____
be required to send an email notice 0--,| /OOOOOOO\
24 hours prior to complete hardware failure! <_/ / /OOOOOOOOOOO\
\ \/OOOOOOOOOOOOOOO\
\ OOOOOOOOOOOOOOOOO|//
Harald Koenig, \/\/\/\/\/\/\/\/\/
Inst.f.Theoret.Astrophysik // / \\ \
koenig@tat.physik.uni-tuebingen.de ^^^^^ ^^^^^
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu