linuxconsole-dev

G550DH + G450DH Framebuffer succes & directfb question.

[Mark H. <Mar...@xs...>]

Hi,


I managed to find a couple of matrox cards (G550 dual head AGP and a G450
dualhead PCI card), and decided to give it a go with ruby.
I did not succeed in setting up a working dualhead X environment yet
(although Xinerama works), but with framebuffer i have everything workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers. There are
still a few issues (I think Helge also mentioned some of them before):

    * upon loading of the matroxfb drivers, all framebuffers show a garbled
      screen.
    * when loading fbcon, the garbled screen cleans up (mostly), but on the
      main console I see a big white rectangle. On none of the consoles I
      can see the login prompt, but it appears after pressing enter key :-)
    * when switching virtual console on the main console with F1-F6 keys,
      the framebuffer setup is somehow messed up and so the main console is
      also shown on the second head. Rerunning the matroxset commands does
      not help, but after rerunning my 'reset_matrox' script which also
      uses fbset to setup the framebuffers, everything works again.

Since the framebuffers are working so well, I decided to try some DirectFB
applications (would like to run freevo or mythtv on my TV and try
the XDirectFB server), however, they don't seem to like ruby because I keep
getting the following error with every application I try to start:

(!) DirectFB/FBDev/vt: FBIOGET_CON2FBMAP failed!
    --> Inappropriate ioctl for device

Is there anything I can do to further investigate? Perhaps I can patch
ruby to add a dummy IOCTL or patch directfb in some way that the IOCTL is
not needed anymore?
Should I take this to the directfb mailinglist, or is it a ruby problem?

The message reminds me of the old con2fbmap tool, which we don't need
anymore since ruby takes care of console to fb mapping :-)


Well, I'm very impressed with ruby's current state on matroxfb, so
good work guys!

Mark.


P.S.
Did any of you have any success setting up dual independent (accelerated) X
servers on the G550? I'm very interested to see your XF86Config!

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Helge H. <hel...@ai...>]

Mark Hurenkamp wrote:

>Hi,
>
>
>I managed to find a couple of matrox cards (G550 dual head AGP and a G450
>dualhead PCI card), and decided to give it a go with ruby.
>I did not succeed in setting up a working dualhead X environment yet
>(although Xinerama works), but with framebuffer i have everything workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers.
>
Did you try modular only, or did you also try compiled-in framebuffer?
I have not been able to run 2.6.3 or later, because of a divide error when
initializing the framebuffers.  I'll try modular now, to see if that 
improves
anything.

> There are
>still a few issues (I think Helge also mentioned some of them before):
>
>    * upon loading of the matroxfb drivers, all framebuffers show a garbled
>      screen.
>  
>
Fortunately, a minor problem. They are usually loaded only once.

>    * when loading fbcon, the garbled screen cleans up (mostly), but on the
>      main console I see a big white rectangle. On none of the consoles I
>      can see the login prompt, but it appears after pressing enter key :-)
>  
>
The fbcons works, they merely start out with some garbage.

>    * when switching virtual console on the main console with F1-F6 keys,
>      the framebuffer setup is somehow messed up and so the main console is
>      also shown on the second head. Rerunning the matroxset commands does
>      not help, but after rerunning my 'reset_matrox' script which also
>      uses fbset to setup the framebuffers, everything works again.
>  
>
Now this is irritating.  It seems to be tied to changing resolution or
video mode. Changing resolution in X messes up the same way.

>Since the framebuffers are working so well, I decided to try some DirectFB
>applications (would like to run freevo or mythtv on my TV and try
>the XDirectFB server), however, they don't seem to like ruby because I keep
>getting the following error with every application I try to start:
>
>(!) DirectFB/FBDev/vt: FBIOGET_CON2FBMAP failed!
>    --> Inappropriate ioctl for device
>
>Is there anything I can do to further investigate? Perhaps I can patch
>  
>
Try on each console, maybe this fails only on some consoles?

>ruby to add a dummy IOCTL or patch directfb in some way that the IOCTL is
>not needed anymore?
>Should I take this to the directfb mailinglist, or is it a ruby problem?
>
>The message reminds me of the old con2fbmap tool, which we don't need
>anymore since ruby takes care of console to fb mapping :-)
>
>
>Well, I'm very impressed with ruby's current state on matroxfb, so
>good work guys!
>
>Mark.
>
>
>P.S.
>Did any of you have any success setting up dual independent (accelerated) X
>servers on the G550? I'm very interested to see your XF86Config!
>
>  
>
I haven't researched this more, but I once ran two accelerated
servers simultaneously.  (Both on the same screen though, but
that's a minor point - clever use of xinerama and or "screen 1"
should make it possible to get on separate screens)

The G550 with two accelerated X servers is fine as long as
you use ordinary 2D.   Run a 3D program on one server and
the other mostly freeze waiting for the 3D stuff to stop.
Two 3D programs simultaneously on separate xservers fail,
the second one simply segfaults.  All this may change with
newer xservers or newer DRI. My attempt was a few months ago.

Helge Hafting

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Aivils <ai...@un...>]

On Thursday 13 May 2004 10:56, Mark Hurenkamp wrote:
> I managed to find a couple of matrox cards (G550 dual head AGP and a G450
> dualhead PCI card), and decided to give it a go with ruby.
> I did not succeed in setting up a working dualhead X environment yet
> (although Xinerama works), but with framebuffer i have everything workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers. There are
> still a few issues (I think Helge also mentioned some of them before):
> 
>     * upon loading of the matroxfb drivers, all framebuffers show a garbled
>       screen.
>     * when loading fbcon, the garbled screen cleans up (mostly), but on the
>       main console I see a big white rectangle. On none of the consoles I
>       can see the login prompt, but it appears after pressing enter key :-)
>     * when switching virtual console on the main console with F1-F6 keys,
>       the framebuffer setup is somehow messed up and so the main console is
>       also shown on the second head. Rerunning the matroxset commands does
>       not help, but after rerunning my 'reset_matrox' script which also
>       uses fbset to setup the framebuffers, everything works again.

Unfortunately i do not understand how hard is fb support,
when i add fbcon. Each fb low level driver slightly differs.
Seems, matrox fb is spiteful.

> Since the framebuffers are working so well, I decided to try some DirectFB
> applications (would like to run freevo or mythtv on my TV and try
> the XDirectFB server), however, they don't seem to like ruby because I keep
> getting the following error with every application I try to start:
> 
> (!) DirectFB/FBDev/vt: FBIOGET_CON2FBMAP failed!
>     --> Inappropriate ioctl for device
> 
> Is there anything I can do to further investigate? Perhaps I can patch
> ruby to add a dummy IOCTL or patch directfb in some way that the IOCTL is
> not needed anymore?
> Should I take this to the directfb mailinglist, or is it a ruby problem?

pure ruby problem.

> The message reminds me of the old con2fbmap tool, which we don't need
> anymore since ruby takes care of console to fb mapping :-)

con2fbmap is to match complicated for current linux-ruby. In abstract
user may use con2fbmap on 3 head (or more) system, when 1st user
uses 2 heads and 2nd user uses 1 head.
Current i agree more simple solution , when each user uses 1 video card
(or head - matrox case), 1 keyboard.
I will revert back original linux con2fb code. Then con2fbmap does dummy
IOCTL, compatible with linus-tree :)

How sane is moving con2fb[ ] array to VT structure ?
Existence of con2fb[ ] breaks all automatic created links between
TTY and fbdev.

Aivils

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Svetoslav S. <sv...@gm...>]

Hi all,

first of all , i've yet ound time to play with ruby-2.6.5,
so here are my experiancies with ruby-2.6.3

> On Thursday 13 May 2004 10:56, Mark Hurenkamp wrote:
> > I managed to find a couple of matrox cards (G550 dual head AGP and a
> G450
> > dualhead PCI card), and decided to give it a go with ruby.
> > I did not succeed in setting up a working dualhead X environment yet
> > (although Xinerama works), but with framebuffer i have everything
> workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers.
There are
> > still a few issues (I think Helge also mentioned some of them before):


i could get it running with fbdev X driver, and i think mga +fb 
on a G550DH, but i needed Aivils hack for VT_SHIFT & 2 Xserver binaries

> > 
> >     * upon loading of the matroxfb drivers, all framebuffers show a
> garbled
> >       screen.
same here

> >     * when loading fbcon, the garbled screen cleans up (mostly), but on
> the
> >       main console I see a big white rectangle. On none of the consoles
> I
> >       can see the login prompt, but it appears after pressing enter key
> :-)

same here
> >     * when switching virtual console on the main console with F1-F6
> keys,
> >       the framebuffer setup is somehow messed up and so the main console
> is
> >       also shown on the second head. Rerunning the matroxset commands
> does

i'm not sure i've seen this

> >       not help, but after rerunning my 'reset_matrox' script which also
> >       uses fbset to setup the framebuffers, everything works again.

 
> Unfortunately i do not understand how hard is fb support,
> when i add fbcon. Each fb low level driver slightly differs.
> Seems, matrox fb is spiteful.

it's very, very spiteful :(

it's just that the maintainer doesn't like the new API,
and doesn't care much about the in kernel driver
and ....
he still ships his own matroxfb driver, which reverts the API 
change almost completely 
:(

but IMO this is the only (?i386?) hardware 
with some support for multi-head fbdev

best,

svetljo

-- 
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgebühr: http://www.gmx.net/dsl

Re: G550DH + G450DH Framebuffer succes & directfb question.

[James S. <jsi...@in...>]

> I managed to find a couple of matrox cards (G550 dual head AGP and a G450
> dualhead PCI card), and decided to give it a go with ruby.
> I did not succeed in setting up a working dualhead X environment yet
> (although Xinerama works), but with framebuffer i have everything workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers. There are
> still a few issues (I think Helge also mentioned some of them before):
> 
>     * upon loading of the matroxfb drivers, all framebuffers show a garbled
>       screen.

This is a bug in the mainline kernel as well. The problem is VGA text 
console is not flushing it text buffers before the framebuffer driver 
takes over and then redraw the buffer. Another problem is some framebuffer
drivers call printk while initializing. When takeover_console is called 
the printk could be called while the hardware is in the in between state 
of vga text mode and graphics mode. The console layer wants to still print 
something to the screen but the hardware is not ready for that.

>     * when loading fbcon, the garbled screen cleans up (mostly), but on the
>       main console I see a big white rectangle. On none of the consoles I
>       can see the login prompt, but it appears after pressing enter key :-)

I have seen that before. The white rectangle is the size of the earlier 
console resolution you just toke over. I see this when I go from vgacon to 
neofb with fbcon. The fbdev layer will most likely end up doing a memory 
clear out before taking over as a fix.

>     * when switching virtual console on the main console with F1-F6 keys,
>       the framebuffer setup is somehow messed up and so the main console is
>       also shown on the second head. Rerunning the matroxset commands does
>       not help, but after rerunning my 'reset_matrox' script which also
>       uses fbset to setup the framebuffers, everything works again.

The matrox is a funny driver. I don't think the matroxset commands where 
meant to be used with the ruby kernel.

> Since the framebuffers are working so well, I decided to try some DirectFB
> applications (would like to run freevo or mythtv on my TV and try
> the XDirectFB server), however, they don't seem to like ruby because I keep
> getting the following error with every application I try to start:
> 
> (!) DirectFB/FBDev/vt: FBIOGET_CON2FBMAP failed!
>     --> Inappropriate ioctl for device

Why is directfb call the console mapping function? That is really bad.

> Is there anything I can do to further investigate? Perhaps I can patch
> ruby to add a dummy IOCTL or patch directfb in some way that the IOCTL is
> not needed anymore?
> Should I take this to the directfb mailinglist, or is it a ruby problem?

I would find out why directfb is playing with con2fb ioctl.

> The message reminds me of the old con2fbmap tool, which we don't need
> anymore since ruby takes care of console to fb mapping :-)
 
> Well, I'm very impressed with ruby's current state on matroxfb, so
> good work guys!
> 
> Mark.
> 
> 
> P.S.
> Did any of you have any success setting up dual independent (accelerated) X
> servers on the G550? I'm very interested to see your XF86Config!

Me too. I like to give a demo of that.

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Svetoslav S. <sv...@gm...>]

> 
> > I managed to find a couple of matrox cards (G550 dual head AGP and a
> G450
> > dualhead PCI card), and decided to give it a go with ruby.
> > I did not succeed in setting up a working dualhead X environment yet
> > (although Xinerama works), but with framebuffer i have everything
> workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers.
There are
> > still a few issues (I think Helge also mentioned some of them before):
> > 
> >     * upon loading of the matroxfb drivers, all framebuffers show a
> garbled
> >       screen.
> 
> This is a bug in the mainline kernel as well. The problem is VGA text 
> console is not flushing it text buffers before the framebuffer driver 
> takes over and then redraw the buffer. Another problem is some framebuffer
> drivers call printk while initializing. When takeover_console is called 
> the printk could be called while the hardware is in the in between state 
> of vga text mode and graphics mode. The console layer wants to still print
> something to the screen but the hardware is not ready for that.
> 
> >     * when loading fbcon, the garbled screen cleans up (mostly), but on
> the
> >       main console I see a big white rectangle. On none of the consoles
> I
> >       can see the login prompt, but it appears after pressing enter key
> :-)
> 
> I have seen that before. The white rectangle is the size of the earlier 
> console resolution you just toke over. I see this when I go from vgacon to
> neofb with fbcon. The fbdev layer will most likely end up doing a memory 
> clear out before taking over as a fix.
> 
> >     * when switching virtual console on the main console with F1-F6
> keys,
> >       the framebuffer setup is somehow messed up and so the main console
> is
> >       also shown on the second head. Rerunning the matroxset commands
> does
> >       not help, but after rerunning my 'reset_matrox' script which also
> >       uses fbset to setup the framebuffers, everything works again.
> 
> The matrox is a funny driver.

agree :(

> I don't think the matroxset commands where 
> meant to be used with the ruby kernel.

there is nothing wrong with matroxset with ruby,
and it's the only way to separate the outputs 
of the two heads of G4x0DH or G550DH
else both fbdevs are mirrored :(

does anyone have a better solution ?

( 
of course it would be great to get independant fbcon 
without matroxset, but i'm not aware of such possibility
)

> > Since the framebuffers are working so well, I decided to try some
> DirectFB
> > applications (would like to run freevo or mythtv on my TV and try
> > the XDirectFB server), however, they don't seem to like ruby because I
> keep
> > getting the following error with every application I try to start:
> > 
> > (!) DirectFB/FBDev/vt: FBIOGET_CON2FBMAP failed!
> >     --> Inappropriate ioctl for device
> 
> Why is directfb call the console mapping function? That is really bad.
> 
> > Is there anything I can do to further investigate? Perhaps I can patch
> > ruby to add a dummy IOCTL or patch directfb in some way that the IOCTL
> is
> > not needed anymore?
> > Should I take this to the directfb mailinglist, or is it a ruby problem?
> 
> I would find out why directfb is playing with con2fb ioctl.
> 
> > The message reminds me of the old con2fbmap tool, which we don't need
> > anymore since ruby takes care of console to fb mapping :-)
>  
> > Well, I'm very impressed with ruby's current state on matroxfb, so
> > good work guys!
> > 
> > Mark.
> > 
> > 
> > P.S.
> > Did any of you have any success setting up dual independent
> (accelerated) X
> > servers on the G550? I'm very interested to see your XF86Config!
> 
> Me too. I like to give a demo of that.

IMO, this is only possible with the Matrox MMS series cards (G200 or G450)
two/ four heads by choice(pocket) , but they are really too
expensive
it's just that the hardware is missing

best,

svetljo

-- 
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgebühr: http://www.gmx.net/dsl

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Helge H. <hel...@ai...>]

Svetoslav Slavtchev wrote:

>>>I managed to find a couple of matrox cards (G550 dual head AGP and a
>>>      
>>>
>>G450
>>    
>>
>>>dualhead PCI card), and decided to give it a go with ruby.
>>>I did not succeed in setting up a working dualhead X environment yet
>>>(although Xinerama works), but with framebuffer i have everything
>>>      
>>>
>>workingnow on kernel 2.6.5, even with modular matroxfb & fbcon drivers.
>>    
>>
>There are
>  
>
>>>still a few issues (I think Helge also mentioned some of them before):
>>>
>>>    * upon loading of the matroxfb drivers, all framebuffers show a
>>>      
>>>
>>garbled
>>    
>>
>>>      screen.
>>>      
>>>
>>This is a bug in the mainline kernel as well. The problem is VGA text 
>>console is not flushing it text buffers before the framebuffer driver 
>>takes over and then redraw the buffer. Another problem is some framebuffer
>>drivers call printk while initializing. When takeover_console is called 
>>the printk could be called while the hardware is in the in between state 
>>of vga text mode and graphics mode. The console layer wants to still print
>>something to the screen but the hardware is not ready for that.
>>
>>    
>>
>>>    * when loading fbcon, the garbled screen cleans up (mostly), but on
>>>      
>>>
>>the
>>    
>>
>>>      main console I see a big white rectangle. On none of the consoles
>>>      
>>>
>>I
>>    
>>
>>>      can see the login prompt, but it appears after pressing enter key
>>>      
>>>
>>:-)
>>
>>I have seen that before. The white rectangle is the size of the earlier 
>>console resolution you just toke over. I see this when I go from vgacon to
>>neofb with fbcon. The fbdev layer will most likely end up doing a memory 
>>clear out before taking over as a fix.
>>
>>    
>>
>>>    * when switching virtual console on the main console with F1-F6
>>>      
>>>
>>keys,
>>    
>>
>>>      the framebuffer setup is somehow messed up and so the main console
>>>      
>>>
>>is
>>    
>>
>>>      also shown on the second head. Rerunning the matroxset commands
>>>      
>>>
>>does
>>    
>>
>>>      not help, but after rerunning my 'reset_matrox' script which also
>>>      uses fbset to setup the framebuffers, everything works again.
>>>      
>>>
>>The matrox is a funny driver.
>>    
>>
>
>agree :(
>
>  
>
>>I don't think the matroxset commands where 
>>meant to be used with the ruby kernel.
>>    
>>
>
>there is nothing wrong with matroxset with ruby,
>and it's the only way to separate the outputs 
>of the two heads of G4x0DH or G550DH
>else both fbdevs are mirrored :(
>
>does anyone have a better solution ?
>  
>
I'm considering patching it so it'll boot up with separate outputs instead
of mirroring by default.  This should ideally be a kernel command
line option, as the current default is useful for those with only one
screen.  (But then, why get a G550 if you use one screen only . . .)

>( 
>of course it would be great to get independant fbcon 
>without matroxset, but i'm not aware of such possibility
>)
>
>  
>

>>>P.S.
>>>Did any of you have any success setting up dual independent
>>>      
>>>
>>(accelerated) X
>>    
>>
>>>servers on the G550? I'm very interested to see your XF86Config!
>>>      
>>>
>>Me too. I like to give a demo of that.
>>    
>>
>
>IMO, this is only possible with the Matrox MMS series cards (G200 or G450)
>two/ four heads by choice(pocket) , but they are really too
>expensive
>it's just that the hardware is missing
>  
>
It is possible on a G550, _if_ you stick to 2D-accelearation only.
Recipe:
Install a ruby kernel, get it going with two independent consoles.
Also connect two independent mice.
Using framebuffers is one way to achieve this.

Install the accelerated X server.
Run two such servers, on separate tty's so they will run simultaneously.
Use two mostly identical XF86Config's, they should specify separate
mice though. At least one of them must use a software mouse cursor,
as the G550 only have one of these.
No need for any prefbusid/isolatedevice because they run on the same
card - and therefore they don't try to disable each other's cards.

This works in practice, the two servers didn't corrupt each others state,
not even with a smp machine that really lets the two drivers execute
simultaneously.

This recipe gives two independent accelerated xservers, using separate
keyboards & mice.  Unfortunately they happen to run on the same
screen, and tend to paint over each other all the time.
I didn't go further than this, but believe you can use xinerama
and simply tell the users to not abuse into each other's screens.
Perhaps it is possible to do better than that - if anyone can
figure out how to run accelerated X on the second head _only_.
(First head only id trivial - it is the standard setup with xinerama off.)

Note that this didn't work well with 3D, last time I tried.
One server using 3D blocks the other xserver 99% of the time, meaning
the other user can't  work efficiently if the first user is using a game.
Even 2D operations get blocked like this.  3D on both servers were
impossible, even though two 3D clients on a single server works fine.

Things seems to happen fast on the DRI front, but I don't really
expect this to work anytime soon.  Virtualizing the 3D accelerator
will make this "easy", but the implied slowdown might not be
desireable. 

Zoltán Böszörményi already showed dual 3D using two independent cards
though, I believe that is the way to go for dual gaming.

Helge Hafting

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Svetoslav S. <sv...@gm...>]

> >
> >>I don't think the matroxset commands where 
> >>meant to be used with the ruby kernel.
> >>    
> >>
> >
> >there is nothing wrong with matroxset with ruby,
> >and it's the only way to separate the outputs 
> >of the two heads of G4x0DH or G550DH
> >else both fbdevs are mirrored :(
> >
> >does anyone have a better solution ?
> >  
> >
> I'm considering patching it so it'll boot up with separate outputs instead
> of mirroring by default.  This should ideally be a kernel command
> line option, as the current default is useful for those with only one
> screen.  (But then, why get a G550 if you use one screen only . . .)

Petr, refused to help me on this issue,
do you have idea how to patch it ?
IMHO if we figure out how to patch it,
it shouldn't be that hard to add the boot option

 
> >( 
> >of course it would be great to get independant fbcon 
> >without matroxset, but i'm not aware of such possibility
> >)
> >
> >  
> >
> 
> >>>P.S.
> >>>Did any of you have any success setting up dual independent
> >>>      
> >>>
> >>(accelerated) X
> >>    
> >>
> >>>servers on the G550? I'm very interested to see your XF86Config!
> >>>      
> >>>
> >>Me too. I like to give a demo of that.
> >>    
> >>
> >
> >IMO, this is only possible with the Matrox MMS series cards (G200 or
> G450)
> >two/ four heads by choice(pocket) , but they are really too
> >expensive
> >it's just that the hardware is missing
> >  
> >
> It is possible on a G550, _if_ you stick to 2D-accelearation only.
> Recipe:
> Install a ruby kernel, get it going with two independent consoles.
> Also connect two independent mice.
> Using framebuffers is one way to achieve this.
> 
> Install the accelerated X server.
> Run two such servers, on separate tty's so they will run simultaneously.
> Use two mostly identical XF86Config's, they should specify separate
> mice though. At least one of them must use a software mouse cursor,
> as the G550 only have one of these.
> No need for any prefbusid/isolatedevice because they run on the same
> card - and therefore they don't try to disable each other's cards.
> 
> This works in practice, the two servers didn't corrupt each others state,
> not even with a smp machine that really lets the two drivers execute
> simultaneously.
> 
> This recipe gives two independent accelerated xservers, using separate
> keyboards & mice.  Unfortunately they happen to run on the same
> screen, and tend to paint over each other all the time.
> I didn't go further than this, but believe you can use xinerama
> and simply tell the users to not abuse into each other's screens.
> Perhaps it is possible to do better than that - if anyone can
> figure out how to run accelerated X on the second head _only_.
> (First head only id trivial - it is the standard setup with xinerama off.)
> 
> Note that this didn't work well with 3D, last time I tried.
> One server using 3D blocks the other xserver 99% of the time, meaning
> the other user can't  work efficiently if the first user is using a game.
> Even 2D operations get blocked like this.  3D on both servers were
> impossible, even though two 3D clients on a single server works fine.

IMHO this is just a happy coinsidence,
and not a proper solution

> Things seems to happen fast on the DRI front, but I don't really
> expect this to work anytime soon.  Virtualizing the 3D accelerator
> will make this "easy", but the implied slowdown might not be
> desireable. 

yep, but IIRC DRI is still single card only :(

> Zoltán Böszörményi already showed dual 3D using two independent cards
> though, I believe that is the way to go for dual gaming.

i had it running on two PC's for a long time,
DRI + NVidia, and IIRC there must be a lot of
users with multiple NVidia cards
it's sort of sad that in order to have
multiple 3D accelerated users you MUST use NVidia
closed source drivers

best,

svetljo

-- 
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgebühr: http://www.gmx.net/dsl

Re: G550DH + G450DH Framebuffer succes & directfb question.

[James S. <jsi...@in...>]

> Note that this didn't work well with 3D, last time I tried.
> One server using 3D blocks the other xserver 99% of the time, meaning
> the other user can't  work efficiently if the first user is using a game.
> Even 2D operations get blocked like this.  3D on both servers were
> impossible, even though two 3D clients on a single server works fine.

:-(
 
> Things seems to happen fast on the DRI front, but I don't really
> expect this to work anytime soon.  Virtualizing the 3D accelerator
> will make this "easy", but the implied slowdown might not be
> desireable. 
> 
> Zoltán Böszörményi already showed dual 3D using two independent cards
> though, I believe that is the way to go for dual gaming.

That was with the Nvidia closed driver?

Re: G550DH + G450DH Framebuffer succes & directfb question.

[Zoltan B. <zb...@fr...>]

James Simmons =EDrta:
>>Note that this didn't work well with 3D, last time I tried.
>>One server using 3D blocks the other xserver 99% of the time, meaning
>>the other user can't  work efficiently if the first user is using a gam=
e.
>>Even 2D operations get blocked like this.  3D on both servers were
>>impossible, even though two 3D clients on a single server works fine.
>=20
>=20
> :-(
>=20
>=20
>>Things seems to happen fast on the DRI front, but I don't really
>>expect this to work anytime soon.  Virtualizing the 3D accelerator
>>will make this "easy", but the implied slowdown might not be
>>desireable.
>>
>>Zolt=E1n B=F6sz=F6rm=E9nyi already showed dual 3D using two independent=
 cards
>>though, I believe that is the way to go for dual gaming.
>=20
>=20
> That was with the Nvidia closed driver?

No, this is with the plain FC1 XFree86 radeon driver.
I had to apply the latest DRI patch from the -mm tree
so it provides /dev/dri/card0 and /dev/dri/card1.
It detects the cards in bus/card order, so the
PCI radeon is card0 and the AGP radeon is card1.
I had to setup the servers in the same order in gdm.conf,
i.e. I have this in dmesg:

[drm] Initialized radeon 1.10.0 20020828 on minor 0: ATI Technologies=20
Inc Radeon RV100 QY [Radeon 7000/VE]
[drm] Initialized radeon 1.10.0 20020828 on minor 1: ATI Technologies=20
Inc RV280 [Radeon 9200 SE]

and this in gdm.conf:

[servers]
0=3DX0
1=3DX1

[server-X0]
name=3DX0 server
command=3D/usr/X11R6/bin/X -deferglyphs 16 -layout "First Head" vt7
flexible=3Dtrue

[server-X1]
name=3DX1 server
command=3D/usr/X11R6/bin/X -deferglyphs 16 -layout "Second Head" vt17
flexible=3Dtrue

and this in XF86Config:

Section "ServerLayout"
         Identifier     "First Head"
         Screen      0  "Screen0" 0 0
         InputDevice    "Mouse1" "CorePointer"
         InputDevice    "Keyboard0" "CoreKeyboard"
         Option      "SingleCard" "true"
EndSection
=20
=20
     Section "ServerLayout"
         Identifier     "Second Head"
         Screen      0  "Screen1" 0 0
         InputDevice    "Mouse0" "CorePointer"
         InputDevice    "Keyboard0" "CoreKeyboard"
         Option      "SingleCard" "true"
EndSection
...
Section "Device"
=20
=20
             Identifier  "Videocard1"
         Driver      "radeon"
         VendorName  "Videocard vendor"
         BoardName   "ATI Radeon 9200SE"
         Option      "AGPMode" "4"
         Option      "dpms" "off"
         BusID       "PCI:1:0:0"
EndSection
Section "Device"
         Identifier  "Videocard0"
         Driver      "radeon"
         VendorName  "Videocard vendor"
         BoardName   "ATI Radeon 7000"
         Option      "dpms" "off"
         BusID       "PCI:0:6:0"
EndSection
Section "Screen"
         Identifier "Screen0"
         Device     "Videocard0"
         Monitor    "Monitor0"
         DefaultDepth     16
         SubSection "Display"
...
EndSection
Section "Screen"
         Identifier "Screen1"
         Device     "Videocard1"
         Monitor    "Monitor0"
         DefaultDepth     24
         SubSection "Display"
...
EndSection

The XFree86-4.3.0 DRI detection gets confused if
the card initialization is reversed. It tries to
set the DRI "private" field to the card's PCI
BusID and it fails if its not the same as the
actual BusID and XFree86 gives up if the device
node is not already opened (busy) and the setting
this private field fails.

Best regards,
Zolt=E1n B=F6sz=F6rm=E9nyi

Re: G550DH + G450DH Framebuffer succes & directfb question.

[James S. <jsi...@in...>]

> > I don't think the matroxset commands where 
> > meant to be used with the ruby kernel.
> 
> there is nothing wrong with matroxset with ruby,
> and it's the only way to separate the outputs 
> of the two heads of G4x0DH or G550DH
> else both fbdevs are mirrored :(
> 
> does anyone have a better solution ?
> 
> ( 
> of course it would be great to get independant fbcon 
> without matroxset, but i'm not aware of such possibility
> )

I haven't fixed it up but I do have a idea. It would also eliminate that 
awful driver init list in fbmem.c.  The idea was to send a event when a 
driver is registered. Then fbcon would recieve that event. Then it would 
look at the con2fb_map in fbcon to see what VCs to map to. At the boot 
command line or via con2fb you can set this mapping. Now if you have the 
case of a dual headed card where one head isn't mapped then it could 
default to mirroring. Fundamentally we need a mirror api for the fbdev 
layer. I'm hoping to introduce that via sysfs.

> IMO, this is only possible with the Matrox MMS series cards (G200 or G450)
> two/ four heads by choice(pocket) , but they are really too
> expensive
> it's just that the hardware is missing

So it is possible to run OpenGl on a Matrox card :-)