linuxconsole-dev

Multi-seat XFree solution under Linux with framebuffers.

[Stian H. <sti...@it...>]

Hi all,

we have just finished writing a patch + a HOWTO for configuring a
multi-seat XFree solution under Linux with framebuffers. 

If any of you at this group would like to test it and give us comments,
we would appreciate that very much.

The HOWTO and patch can be found at
http://www.itsopen.net/projects/x-hack/

Have a nice (testing) day.


Regards

Stian Hartviksen (sha...@it...)
ItsOpen AS

Re: Multi-seat XFree solution under Linux with framebuffers.

[<shs...@st...>]

Quoting Stian Hartviksen <sti...@it...>:

> Hi all,
> 
> we have just finished writing a patch + a HOWTO for configuring a
> multi-seat XFree solution under Linux with framebuffers. 
> 
> If any of you at this group would like to test it and give us
> comments,
> we would appreciate that very much.
> 
> The HOWTO and patch can be found at
> http://www.itsopen.net/projects/x-hack/
> 
> Have a nice (testing) day.
> 
> 
> Regards
> 
> Stian Hartviksen (sha...@it...)
> ItsOpen AS

Hi it may sound a bit like critics,
but what did you changed in the Miguel Freitas patch, How-To ...
i read your How-To and i don't seem to find any difference,
except that you tottaly disable VT switching and 
the PCI resourse management (access disable functions),
the last one wasn't seen good from the XFree people.

Best,

svetljo 
 

Re: Multi-seat XFree solution under Linux with framebuffers.

[Stian H. <sti...@it...>]

On Wed, 2003-03-12 at 11:40, shs...@st...
wrote:

> 
> Hi it may sound a bit like critics,
> but what did you changed in the Miguel Freitas patch, How-To ...
> i read your How-To and i don't seem to find any difference,
> except that you tottaly disable VT switching and 
> the PCI resourse management (access disable functions),
> the last one wasn't seen good from the XFree people.
> 
> Best,
> 
> svetljo 
>  

Hi, Svetljo.

There are not that many changes from Miguel Freitas, but the main
difference from our point of view is that we now can use Framebuffer
drivers with Multi-seat.

The patch from Miguel Freitas was removing all the framebuffer stuff,
and we have just removed some parts of it instead. We had to disable the
VT Switching and PCI Resource management to get more than one active
X-Server at the same time.

If you know of any other way to do this (in a better way), please send
me a note!


WBR Stian Hartviksen

Re: Multi-seat XFree solution under Linux with framebuffers.

[<shs...@st...>]

Quoting Stian Hartviksen <sti...@it...>:

> On Wed, 2003-03-12 at 11:40, shs...@st...
> wrote:
> 
> > 
> > Hi it may sound a bit like critics,
> > but what did you changed in the Miguel Freitas patch, How-To ...
> > i read your How-To and i don't seem to find any difference,
> > except that you tottaly disable VT switching and 
> > the PCI resourse management (access disable functions),
> > the last one wasn't seen good from the XFree people.
> > 
> > Best,
> > 
> > svetljo 
> >  
> 
> Hi, Svetljo.
> 
> There are not that many changes from Miguel Freitas, but the main
> difference from our point of view is that we now can use Framebuffer
> drivers with Multi-seat.
> 
> The patch from Miguel Freitas was removing all the framebuffer stuff,

i don't think so:
have you looked at the explanation about G450,
it's useing just FB's
http://cambuca.ldhs.cetuc.puc-rio.br/multiuser/g450.html

i got this way a Matrox G550DH for 2 X users,
but useing plain FB X drivers means no accel, so ...
 
> and we have just removed some parts of it instead. We had to disable
> the
> VT Switching and PCI Resource management to get more than one active
> X-Server at the same time.
> 
> If you know of any other way to do this (in a better way), please send
> me a note!

i think the way Miguel used for the PCI resource managment was better,
and i don't know a better way. This method( the one from Miguel ) was found by 
some NetBSD users, and is used by backstreet ruby: http://startx.times.lv/

regards,

svetljo

Re: Multi-seat XFree solution under Linux with framebuffers.

[Stian H. <sti...@it...>]

Hi Svetljo, and thank you for you input!


> > The patch from Miguel Freitas was removing all the framebuffer stuff,
> i don't think so:
> have you looked at the explanation about G450,
> it's useing just FB's
> http://cambuca.ldhs.cetuc.puc-rio.br/multiuser/g450.html
> 
> i got this way a Matrox G550DH for 2 X users,
> but useing plain FB X drivers means no accel, so ...

Yes, we have tried this solution, but we have not been able to get this
patch to work with other framebuffer drivers than Matrox (Matrox seems
to be a very "nice" driver). The problem occurred when we needed support
for more than one card/manufacturer (eg. Nvidia, ATI, Matrox, sis). Our
main environment are 4 ATI cards with 8 output on each card ( A total of
32 outputs from one PC - each with its own mouse and keyboard ), and the
solution from Miguel Freitas did not work in our environment.  

> > If you know of any other way to do this (in a better way), please send
> > me a note!
> 
> i think the way Miguel used for the PCI resource managment was better,
> and i don't know a better way. This method( the one from Miguel ) was found by 
> some NetBSD users, and is used by backstreet ruby: http://startx.times.lv/

I will look into this once again, but I know we had some problems with
only one X-session showing at the time (and the rest went blank or
"sleeping") when we used the patch from Freitas.

We have also tried the backstreet ruby solution, but this was not
supporting the kernel version that we needed at the time when we tried
it. Maybe it will work now ? (Kernel 2.4.21-pre5 or later)


WBR Stian Hartviksen

Re: Multi-seat XFree solution under Linux with framebuffers.

[<shs...@st...>]

Hi,

> > may be you could try the Matrox multy series,
> ;-) Sorry, but we have developed a new ATI card (with 8 outputs), and
> need to get the X servers to run on these cards.
> 
> > they should have released a 4 headed card with the G450 chip,
> > and if they made it like the previous series,
> > you may be will be able to run them in both Miguels & backstreet ruby
> way,
> > so you'll be able to get 16-20-24 heads/users 4-5 pci + 1 agp x 4
> heads,
> > and if you have luck you'll be able to get it working with normal
> drivers,
> > not through FB , so you could use Xvideo, DRI ....
> 
> Main problem is that I also need the FB support for our
> software/applications.....

i was just saying that if you have luck you can get the normal drivers work,
but the fb driver should work without a problem,
are you working on a kind of set-top-box, do you need the fb driver in order to 
use X over video out  

i was also thinking that if you get backstreet ruby in workable for you state,
it would be easier to setup the X config files, mostly the keyboard input 
section, as you don't need to mess with the event devices,
with your current solution you should find wich event devices are mices, 
keyboards, additional multimedia intefaces of some keyboards, with ruby you'll 
have to consider only the last two, also it should be fairly easy to get only 
the normal keyboard interfaces to bind to a VT, even if some multimedia 
keyboard interface has registered VT you can fix it trough /proc.
This should be also doable in kernel:
i think all the normal keyboard interfaces register with PHYS= USBxxx0,
and the multimedia interfaces register with PHYS=USBxxx1 , so if you modiffy 
the kernel to bind VTs only to keyboards with PHYS=USBxxx0 you get VTs bind to 
all keyboards -- no need to modiffy X config files if some event device change 
it's number.
 
if i'm not misstaken upon replug of a usb event device it's possible,
that it's event number change -- you'll have to modify the coresponding 
XFConfig to get keyboard back, which shouldn't be needed under ruby, as ruby 
doesn't use the event device files.
   
best,

svetljo

Re: Multi-seat XFree solution under Linux with framebuffers.

[<shs...@st...>]

Quoting shs...@st...:

> Hi,
> 
> > > may be you could try the Matrox multy series,
> > ;-) Sorry, but we have developed a new ATI card (with 8 outputs),
> and
> > need to get the X servers to run on these cards.
> > 
> > > they should have released a 4 headed card with the G450 chip,
> > > and if they made it like the previous series,
> > > you may be will be able to run them in both Miguels & backstreet
> ruby
> > way,
> > > so you'll be able to get 16-20-24 heads/users 4-5 pci + 1 agp x 4
> > heads,
> > > and if you have luck you'll be able to get it working with normal
> > drivers,
> > > not through FB , so you could use Xvideo, DRI ....
> > 
> > Main problem is that I also need the FB support for our
> > software/applications.....

OOPS,
totaly forgot that ruby currently doesn't have FB support,
so unless you convince James Simmons to get back to ruby,
you can not use it if you need FB

sorry,

svetljo

Re: Multi-seat XFree solution under Linux with framebuffers.

[<shs...@st...>]

Hi,

> for more than one card/manufacturer (eg. Nvidia, ATI, Matrox, sis).
what do you mean by that  ^^^^^^

> Our
> main environment are 4 ATI cards with 8 output on each card ( A total
> of
> 32 outputs from one PC - each with its own mouse and keyboard ), and
> the
> solution from Miguel Freitas did not work in our environment.  

can you get more then one fb device for cards other then Matrox,
if the card has several heads but in one chip, i think you wont be able 
to get them work , as they register a single fb device, i think can get all 
heads only under windows or with special closed source drivers driven by a 
single X server. I was thinking that most of the cards act to the OS as a 
single head, but the card itself drives/ separates /manage more monitors.
(i'm having the Nvidia cards in mind, i don't think you can get them work as 
dual head with the open source driver, but i might be misstaken) 


> > > If you know of any other way to do this (in a better way), please
> send
> > > me a note!
> > 
> > i think the way Miguel used for the PCI resource managment was
> better,
> > and i don't know a better way. This method( the one from Miguel ) was
> found by 
> > some NetBSD users, and is used by backstreet ruby:
> http://startx.times.lv/
> 
> I will look into this once again, but I know we had some problems with
> only one X-session showing at the time (and the rest went blank or
> "sleeping") when we used the patch from Freitas.

i know, some cards doesn't play nice, the explanation was that they need to 
disable all other PCI resources in order to initialise.

i had a found some discussion about the problem, but i don't have right now the 
link, i'll post it later. 

> We have also tried the backstreet ruby solution, but this was not
> supporting the kernel version that we needed at the time when we tried
> it. Maybe it will work now ? (Kernel 2.4.21-pre5 or later)
> 

it's up to 2.4.20, do you really need a bleeding new kernel?
you can try to sync it to 2.4.21-pre5.

best,

svetljo

Re: Multi-seat XFree solution under Linux with framebuffers.

[Stian H. <sti...@it...>]

Hi, Svetljo, and sorry about the delay in answering you mail.

> > for more than one card/manufacturer (eg. Nvidia, ATI, Matrox, sis).
> what do you mean by that  ^^^^^^

What I mean is that I want to ba able to use one Nvidia card as the main
card, one ATI on the first X-Server and one Matrox card on the second,
and maybe a SIS on a third one.


> can you get more then one fb device for cards other then Matrox,

I think this will be ok, yes. The driver is hopefully finished in about
14 days for ATI (For our specific card).

> if the card has several heads but in one chip, i think you wont be able 
> to get them work , as they register a single fb device, i think can get all 
> heads only under windows or with special closed source drivers driven by a 

The solution on our card is 4 ATI Mobility chips with "dual head"/ two
independant outputs functionality. This makes a total of 8 output from
one card.

> single X server. I was thinking that most of the cards act to the OS as a 
> single head, but the card itself drives/ separates /manage more monitors.
> (i'm having the Nvidia cards in mind, i don't think you can get them work as 
> dual head with the open source driver, but i might be misstaken) 

We have not been able to get dualhead functionality for the Nvidia
cards. (We are not able to have more than one Nvidia card in the machine
at the same time either. This should be a driver problem, and I have not
had the time or knowledge to examine this.)

 
> > I will look into this once again, but I know we had some problems with
> > only one X-session showing at the time (and the rest went blank or
> > "sleeping") when we used the patch from Freitas.
> 
> i know, some cards doesn't play nice, the explanation was that they need to 
> disable all other PCI resources in order to initialise.
> 
> i had a found some discussion about the problem, but i don't have right now the 
> link, i'll post it later. 

Thank you !


> > We have also tried the backstreet ruby solution, but this was not
> > supporting the kernel version that we needed at the time when we tried
> > it. Maybe it will work now ? (Kernel 2.4.21-pre5 or later)
> > 
> 
> it's up to 2.4.20, do you really need a bleeding new kernel?

I'm sorry to say, yes. We need some USB functionality that has been
taken into the new 2.4.21-pre5 (We are sending 32 mouses and keyboards
over a network protocol, and need them to be treated as USB keyboards
and mouses at the target machine.)

> you can try to sync it to 2.4.21-pre5.

I will try to do this in the weekend if I get the time to do it....


> best,
> 
> svetljo


WBR Stian Hartviksen