two aty128 frame buffers?

two aty128 frame buffers?

[Jeffery von Ronne]

Does anyone know how hard it would be to hack aty128fb to support a
second Rage 128?  Or am I better off trying to get the X server to talk
directly to the second card?

I have two PCI Rage 128's in my G4, which I'd like to use in a
dual-headed Xinerama set up.

So far I've only been able to get the X server (4.1.0 and 4.2.0) to
work through framebuffer driven cards (X's ati/r128 drivers with aty128fb
and X's fbdev with offb).

When I've tried to tell X to talk directly to the unmanaged video card it
complains about some unresolved drm symbols (they're different for 4.1
and 4.2, for 4.1 its:  "Symbol drmScatterGAtherAlooc from module
/usr/X11R6/lib/modules/drivers/r128_drv.o").  I've tried with and without
drm supporting kernels.

Does anyone have any suggestions?

Thanks,
--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Kaoru Fukui]

On  4 Feb, Jeffery von Ronne wrote:
>
> Does anyone know how hard it would be to hack aty128fb to support a
> second Rage 128?  Or am I better off trying to get the X server to talk
> directly to the second card?
>
> I have two PCI Rage 128's in my G4, which I'd like to use in a
> dual-headed Xinerama set up.
>
> So far I've only been able to get the X server (4.1.0 and 4.2.0) to
> work through framebuffer driven cards (X's ati/r128 drivers with aty128fb
> and X's fbdev with offb).
>
> When I've tried to tell X to talk directly to the unmanaged video card it
> complains about some unresolved drm symbols (they're different for 4.1
> and 4.2, for 4.1 its:  "Symbol drmScatterGAtherAlooc from module
> /usr/X11R6/lib/modules/drivers/r128_drv.o").  I've tried with and without
> drm supporting kernels.
>
> Does anyone have any suggestions?

Add drm in Section Module

Section "Module"
	Load  "drm"
EndSection

Kaoru


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Tue, Feb 05, 2002 at 02:21:27PM +0900, Kaoru Fukui wrote:
> Add drm in Section Module

Thanks.  I should have been able to figure that out.

Unfortunately, that was not my real problem.  I guess I forgot to mention
that it crashes with a "signal 11" after the error message; now it crashes
with a signal 11 and no error message.

If anyone is interested in looking at XF86Config-4 and XFree86.0.log,
they are at <URL: http://www.ics.uci.edu/~jronne/tmp/>.

--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Michel Dänzer]

[-- Attachment #1: Type: text/plain, Size: 614 bytes --]

On Die, 2002-02-05 at 07:22, Jeffery von Ronne wrote:
> 
> On Tue, Feb 05, 2002 at 02:21:27PM +0900, Kaoru Fukui wrote:
> > Add drm in Section Module
> 
> Thanks.  I should have been able to figure that out.
> 
> Unfortunately, that was not my real problem.  I guess I forgot to mention
> that it crashes with a "signal 11" after the error message; now it crashes
> with a signal 11 and no error message.

You need a hack like the attached one.


-- 
Earthling Michel Dänzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member   /  CS student, Free Software enthusiast

[-- Attachment #2: r128-vgaHW.diff --]
[-- Type: text/plain, Size: 1110 bytes --]

Index: r128_driver.c
===================================================================
RCS file: /cvs/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_driver.c,v
retrieving revision 1.57
diff -u -r1.57 r128_driver.c
--- r128_driver.c	2002/01/04 21:22:26	1.57
+++ r128_driver.c	2002/02/05 09:24:32
@@ -186,6 +186,7 @@
     "vgaHWGetIndex",
     "vgaHWLock",
     "vgaHWRestore",
+    "vgaHWGetIOBase",
     "vgaHWSave",
     "vgaHWUnlock",
     NULL
@@ -1793,6 +1794,8 @@
 	return FALSE;
     }
 
+    vgaHWGetIOBase(VGAHWPTR(pScrn));
+
     info->PciInfo      = xf86GetPciInfoForEntity(info->pEnt->index);
     info->PciTag       = pciTag(info->PciInfo->bus,
 				info->PciInfo->device,
@@ -2755,7 +2758,10 @@
 	return;
     }
     vgaHWUnlock(hwp);
-    vgaHWSave(pScrn, &hwp->SavedReg, VGA_SR_ALL); /* save mode, fonts, cmap */
+    if (info->VBIOS)
+        vgaHWSave(pScrn, &hwp->SavedReg, VGA_SR_ALL); /* save mode, fonts, cmap */
+    else
+        vgaHWSave(pScrn, &hwp->SavedReg, VGA_SR_MODE); /* save mode */
     vgaHWLock(hwp);
 
     R128SaveMode(pScrn, save);

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Tue, Feb 05, 2002 at 10:28:33AM +0100, Michel Dänzer wrote:
> > Unfortunately, that was not my real problem.  I guess I forgot to mention
> > that it crashes with a "signal 11" after the error message; now it crashes
> > with a signal 11 and no error message.
>
> You need a hack like the attached one.

I tried applying your hack to a virgin XF 4.2.0 source tree, and
I get a signal 11 crash.  The last output shows the X server loading
vgahw.  Are there any other patches I should be applying?

My server log and configuration are at
<URL: http://www.ics.uci.edu/~jronne/2-XFree86.0.log> and
<URL: http://www.ics.uci.edu/~jronne/2-XF86Config-4>.

Thanks,
--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Michel Dänzer]

On Mit, 2002-02-06 at 00:16, Jeffery von Ronne wrote:
> On Tue, Feb 05, 2002 at 10:28:33AM +0100, Michel Dänzer wrote:
> > > Unfortunately, that was not my real problem.  I guess I forgot to mention
> > > that it crashes with a "signal 11" after the error message; now it crashes
> > > with a signal 11 and no error message.
> >
> > You need a hack like the attached one.
>
> I tried applying your hack to a virgin XF 4.2.0 source tree, and
> I get a signal 11 crash.  The last output shows the X server loading
> vgahw.

Hmm, isn't info->VBIOS == NULL? If not, enforce the other route in that
if construct. :)


--
Earthling Michel Dänzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member   /  CS student, Free Software enthusiast

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Wed, Feb 06, 2002 at 01:08:28AM +0100, Michel Dänzer wrote:
> > I tried applying your hack to a virgin XF 4.2.0 source tree, and
> > I get a signal 11 crash.  The last output shows the X server loading
> > vgahw.

I strated sprinkling the code with R128_TRACE messages, and as near as I can
tell, the vgaHWGetIOBase is crashing.

--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Michel Dänzer]

On Don, 2002-02-07 at 00:37, Jeffery von Ronne wrote:
> On Wed, Feb 06, 2002 at 01:08:28AM +0100, Michel Dänzer wrote:
> > > I tried applying your hack to a virgin XF 4.2.0 source tree, and
> > > I get a signal 11 crash.  The last output shows the X server loading
> > > vgahw.
>
> I strated sprinkling the code with R128_TRACE messages, and as near as I can
> tell, the vgaHWGetIOBase is crashing.

Works here with a single chip... can you send me your XF86Config and X
server log privately?


--
Earthling Michel Dänzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member   /  CS student, Free Software enthusiast

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Ani Joshi]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 811 bytes --]


On Wed, 6 Feb 2002, Jeffery von Ronne wrote:

>
> On Wed, Feb 06, 2002 at 01:08:28AM +0100, Michel Dänzer wrote:
> > > I tried applying your hack to a virgin XF 4.2.0 source tree, and
> > > I get a signal 11 crash.  The last output shows the X server loading
> > > vgahw.
>
> I strated sprinkling the code with R128_TRACE messages, and as near as I can
> tell, the vgaHWGetIOBase is crashing.

You can't use 2 cards which require ISA IO (vgaHW uses ISA IO on rage128)
right now, because of the multiple busses on the G4 machines.  You can
comment out all the vgaHW calls and that will probably work, but without
doing that it will always crash on any ISA IO operation.  The reason is
because your second bus's ISA IO isn't mapped.


ani


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Derrik Pates]

On Wed, 6 Feb 2002, Ani Joshi wrote:

> You can't use 2 cards which require ISA IO (vgaHW uses ISA IO on rage128)
> right now, because of the multiple busses on the G4 machines.  You can
> comment out all the vgaHW calls and that will probably work, but without
> doing that it will always crash on any ISA IO operation.  The reason is
> because your second bus's ISA IO isn't mapped.

The issue is the same with the G3 tower - the video card is by default in
the 32-bit, 66 MHz PCI slot, which is a logically separate bus from the
64-bit, 33 MHz bus that the other 3 slots are part of. What sort of mods
will it take to make that work, anyway?

I've discovered that putting both cards in bus 0 (64-bit, 33 MHz bus) gets
both the cards going, though - and well, that works for me...

Derrik Pates      |   Sysadmin, Douglas School   |    #linuxOS on EFnet
dpates@dsdk12.net |     District (dsdk12.net)    |    #linuxOS on OPN


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Ani Joshi]

On Thu, 7 Feb 2002, Derrik Pates wrote:

> The issue is the same with the G3 tower - the video card is by default in
> the 32-bit, 66 MHz PCI slot, which is a logically separate bus from the
> 64-bit, 33 MHz bus that the other 3 slots are part of. What sort of mods
> will it take to make that work, anyway?

There is work now in X to get this working, it requires a layer of PCI
abstraction which is pretty raw at the moment.  Hopefully in a few months
something solid should come out.  However, since we're discussing Rage128
cards in Mac's here, it can be said that the vga IO that is being done is
totally unnecessary, as the Rage128 does not require a vga mode to be
saved or restored on Macs.  So totally disabling the vga IO if you are
just using those 2 cards on your machine can be a valid solution (for your
situation, obviously not an X solution).

> I've discovered that putting both cards in bus 0 (64-bit, 33 MHz bus) gets
> both the cards going, though - and well, that works for me...

Yes, unfortunately though people on AGP G4's tend to use an AGP and a PCI
card.  I suppose they could do what you are doing and use the bottom 2 (or
however it works on G4s) PCI cards if they have 2 PCI boards.


ani


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Thu, Feb 07, 2002 at 11:26:43AM -0800, Ani Joshi wrote:
> Yes, unfortunately though people on AGP G4's tend to use an AGP and a PCI
> card.  I suppose they could do what you are doing and use the bottom 2 (or
> however it works on G4s) PCI cards if they have 2 PCI boards.

This, is actually--I think--the setup I have: two PCI R128s on an AGP
G4.  cat /proc/pci shows an AGP bus 0, PCI bus 1, a PCI bus 2 (which
has all of the real devices, including my two R128 video cards) and
most of the internal PCI stuff, and a PCI bus 3.

FYI, on bootup I get some PCI errors related to the video cards:
  PCI:02:03.0: Resource 0: 84000000-87ffffff (f=1208)
  PCI: Cannot allocate resource region 0 of device 02:03.0
  PCI:  parent is c03c0068: 80000000-8fffffff (f=200)
  PCI:02:03.0: Resource 2: 80088000-8008bfff (f=200)
  PCI:02:07.0: Resource 0: 80000000-8007ffff (f=200)
  PCI:02:02.0: Resource 0: 88000000-8bffffff (f=1208)
  PCI: Cannot allocate resource region 0 of device 02:02.0
  PCI:  parent is c03c0068: 80000000-8fffffff (f=200)
  PCI:02:02.0: Resource 1: 00001000-000010ff (f=101)
  PCI:02:02.0: Resource 2: 8008c000-8008ffff (f=200)
  PCI:02:03.0: Resource 1: 00001000-000010ff (f=101)
  PCI: Cannot allocate resource region 1 of device 02:03.0
  PCI:  parent is c261b924: 00001000-00001fff (f=101)
  PCI:02:08.0: Resource 0: 80083000-80083fff (f=200)
  PCI:02:09.0: Resource 0: 80082000-80082fff (f=200)
  PCI:02:0a.0: Resource 0: 80081000-800817ff (f=200)
  PCI:02:0a.0: Resource 1: 80084000-80087fff (f=200)
  PCI:02:0b.0: Resource 0: 00001000-0000107f (f=101)
  PCI: Cannot allocate resource region 0 of device 02:0b.0
  PCI:  parent is c261b924: 00001000-00001fff (f=101)
  PCI:02:0b.0: Resource 1: 80080000-800803ff (f=200)
02:02.0 is my /dev/fb0 ATI R128
02:03.0 is my /dev/fb1 ATI R128
02:0b.0 is my tulip ethernet chip

PS. My full /proc/pci and dmesg are at http://www.ics.uci.edu/~jronne/tmp

--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Timothy A. Seufert]

At 11:29 AM -0700 2/7/02, Derrik Pates wrote:
>On Wed, 6 Feb 2002, Ani Joshi wrote:
>
>>  You can't use 2 cards which require ISA IO (vgaHW uses ISA IO on rage128)
>>  right now, because of the multiple busses on the G4 machines.  You can
>>  comment out all the vgaHW calls and that will probably work, but without
>>  doing that it will always crash on any ISA IO operation.  The reason is
>>  because your second bus's ISA IO isn't mapped.
>
>The issue is the same with the G3 tower - the video card is by default in
>the 32-bit, 66 MHz PCI slot, which is a logically separate bus from the
>64-bit, 33 MHz bus that the other 3 slots are part of.

Actually it's not the same issue.  The blue&white G3 tower really has
just one logical PCI bus.  The host bridge interfaces the processor
to the 66 MHz bus, and the 33 MHz bus is bridged to the 66 MHz bus
via a PCI-to-PCI bridge IC.  I'm pretty sure ISA I/O is already
supposed to work for busses behind bridges.

The problem in the G4s is that Apple's Uni-N host bridge has three
logically distinct PCI busses which are peers rather than
subordinates.  Each of them thinks it is PCI bus #0 -- the namespaces
are separate.  The Linux kernel APIs for ISA I/O don't account for
the possibility of multiple PCI roots.
--
Tim Seufert

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Mon, Feb 04, 2002 at 03:46:26PM -0800, Jeffery von Ronne wrote:
> Does anyone know how hard it would be to hack aty128fb to support a
> second Rage 128?  Or am I better off trying to get the X server to talk
> directly to the second card?

FYI, hacking the kernel's aty128fb driver---by removing the return 0;
in aty128fb_init's PCI probe loop---I can get an aty128fb for both cards.
So far, I've only gotten X's fbdev driver to work with it.  I looks closer
into why r128/UseFB isn't working some other time.

Unfortunately /dev/fb1 (which is the one open firmware uses, BTW) has a couple
glitches.

First, it only seems to work in 8bpp: in other modes the screen splits
into three parts: a top-left, top-right, and bottom.  The bottom just
has garbage, and the top-left and top-right seem to be mirrored
minature of a garbaged version what the screen is supposed to contain.

Second, it seems to have trouble erasing things (e.g. instead of putting
up the X background grid when I move a window or even initially.)

--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Jeffery von Ronne]

On Thu, Feb 07, 2002 at 11:55:59AM -0800, Jeffery von Ronne wrote:
> FYI, hacking the kernel's aty128fb driver---by removing the return 0;
> in aty128fb_init's PCI probe loop---I can get an aty128fb for both cards.
> So far, I've only gotten X's fbdev driver to work with it.  I looks closer
> into why r128/UseFB isn't working some other time.

I can get X's r128_drv/UseFB driver to work with /dev/fb0, but attempting
to use anything other than X's fbdev with /dev/fb1 sig-11's (it seems to
be the same no matter what driver /dev/fb0 is using).

> Second, it seems to have trouble erasing things (e.g. instead of putting
> up the X background grid when I move a window or even initially.)

This actually happens on both /dev/fb0 and /dev/fb1 when X's fbdev driver
is used.  It seems that it often just forgets to copy its shadow framebuffer
to the real framebuffer.  Setting ShadowFB to false, seemed to fix that.

PS An XF86Config-4 and XFree86.0.log from when I tried to use fbdev on /dev/fb0
   and r128 on /dev/fb1 is at http://www.ics.uci.edu/~jronne/tmp

--
Jeffery von Ronne <jronne@ics.uci.edu>
Graduate Student
Information and Computer Science
University of California, Irvine


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: two aty128 frame buffers?

[Michel Dänzer]

On Fre, 2002-02-08 at 02:15, Jeffery von Ronne wrote:
>
> On Thu, Feb 07, 2002 at 11:55:59AM -0800, Jeffery von Ronne wrote:
> > FYI, hacking the kernel's aty128fb driver---by removing the return 0;
> > in aty128fb_init's PCI probe loop---I can get an aty128fb for both cards.
> > So far, I've only gotten X's fbdev driver to work with it.  I looks closer
> > into why r128/UseFB isn't working some other time.
>
> I can get X's r128_drv/UseFB driver to work with /dev/fb0, but attempting
> to use anything other than X's fbdev with /dev/fb1 sig-11's (it seems to
> be the same no matter what driver /dev/fb0 is using).

The r128 driver lacks Option "fbdev" to choose the framebuffer device to
use, so it will always try to use /dev/fb0. Should be easy to add
looking at the fbdev driver.

> > Second, it seems to have trouble erasing things (e.g. instead of putting
> > up the X background grid when I move a window or even initially.)
>
> This actually happens on both /dev/fb0 and /dev/fb1 when X's fbdev driver
> is used.  It seems that it often just forgets to copy its shadow framebuffer
> to the real framebuffer.  Setting ShadowFB to false, seemed to fix that.

Bummer, sounds like there's a problem with the mishadow code.

> PS An XF86Config-4 and XFree86.0.log from when I tried to use fbdev on /dev/fb0
>    and r128 on /dev/fb1 is at http://www.ics.uci.edu/~jronne/tmp

I guess the segfault is due to both drivers trying to use /dev/fb0, but
keep in mind that vgaHW only works for bus 0 unless you change that in
lnx_video.c.


--
Earthling Michel Dänzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member   /  CS student, Free Software enthusiast

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/