reboot weirdness - problem identified

reboot weirdness - problem identified

[Paul Dorman]

Hi all again,

well, it turns out that the server reboots when gdm tries to start. I have 
RageXL adapters, so if anyone has these working successfully I'd love some 
help. How did they go about configuring the correct kernel features?

Thanks,
Paul.


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Ian Pratt]

> well, it turns out that the server reboots when gdm tries to start. I have 
> RageXL adapters, so if anyone has these working successfully I'd love some 
> help. How did they go about configuring the correct kernel features?

What happens if you boot to a boot to a console and then start X
manually without any of the gnome stuff?

Are you trying to compile in agpgart, drm or other fbdev
components into the kernel? I'd certainly recommend trying to
avoid having any of these in the kernel -- most X server drivers
will work fine without them, though with reduced performance
[*]. What happens if you just use the standard X "ati" driver?

There was an email thread recently where attempts were made to
make agpgart/drm work, but it ended inconclusively. We don't have
any hardware that actually needs agpgart or any of the fbdev
components so we haven't been able to test this. You might like
to try the patches in the linux-2.6.8.1-patches directory in the
repo.

Part of the problem is that these drivers have only ever been
used on x86 hardware, so the authors have been rather 'fast and
loose' with their use of various kernel functions (such as
assuming bus and physical addresses are the same).  Most other
Linux drivers are rather more careful as they have to work on
PPC, Alpha etc, and Xen relies on this.

agpgart.patch and drm.patch clean up some obvious "bugs", but I
suspect more fixes will be required. Someone who actually has
some of this hardware will need to fix it.

Ian

[*] I believe the Intel i815 driver is an exception to this; it
requires agpgart.


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Paul Dorman]

Just prior to your email I had tried to start X from the command line after 
taking gdm out of the startup. Again with the spontaneous reboot. I'm not 
really concerned about graphical performance at all, I just prefer working in 
a graphical environment if I can. I'm also considering a case whereby our web 
developers may one day work with Xen VMs on their workstations (thus saving 
our precious server resources) which would be transfered easily to the 
servers when they are ready), in which case having X play nice would be 
crucial.

If anyone has a machine with a RageXL adapter or another one that uses the ati 
xfree86 driver, I'd love a copy of their config file, and to know what they 
had to do to get it all working with the Xen0 kernel.

In the meantime I'm having a go with the agp and drm patches to see if that 
improves the situation.

Thanks all for your time. Regards,
Paul


On Wednesday 22 September 2004 11:33 am, Ian Pratt wrote:
> > well, it turns out that the server reboots when gdm tries to start. I
> > have RageXL adapters, so if anyone has these working successfully I'd
> > love some help. How did they go about configuring the correct kernel
> > features?
>
> What happens if you boot to a boot to a console and then start X
> manually without any of the gnome stuff?
>
> Are you trying to compile in agpgart, drm or other fbdev
> components into the kernel? I'd certainly recommend trying to
> avoid having any of these in the kernel -- most X server drivers
> will work fine without them, though with reduced performance
> [*]. What happens if you just use the standard X "ati" driver?
>
> There was an email thread recently where attempts were made to
> make agpgart/drm work, but it ended inconclusively. We don't have
> any hardware that actually needs agpgart or any of the fbdev
> components so we haven't been able to test this. You might like
> to try the patches in the linux-2.6.8.1-patches directory in the
> repo.
>
> Part of the problem is that these drivers have only ever been
> used on x86 hardware, so the authors have been rather 'fast and
> loose' with their use of various kernel functions (such as
> assuming bus and physical addresses are the same).  Most other
> Linux drivers are rather more careful as they have to work on
> PPC, Alpha etc, and Xen relies on this.
>
> agpgart.patch and drm.patch clean up some obvious "bugs", but I
> suspect more fixes will be required. Someone who actually has
> some of this hardware will need to fix it.
>
> Ian
>
> [*] I believe the Intel i815 driver is an exception to this; it
> requires agpgart.
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
> Project Admins to receive an Apple iPod Mini FREE for your judgement on
> who ports your project to Linux PPC the best. Sponsored by IBM.
> Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Paul Dorman]

I'm still having no luck with this, though I think my expectations are pretty 
reasonable at this time.  I used xf86config to generate a new minimal 
XF86Config-4 file. The key components of it are:

Load "dbe"
Option "omit xfree86-dga"
Load "type1"
Load "speedo"
Driver "vga"
Driver "vesa"

[note I've pulled these out of context, as I'm sure most of you know where 
they go]. 

Even with this the server spontaneously reboots if I try to start x, with 
nothing in the log files. I'm using a fresh Xen build with pristine kernel 
(no agp or drm patches). Starting X from a different kernel works fine.

I'm beginning to think it's something stupid that I've overlooked....

Paul


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Ian Pratt]

> I'm still having no luck with this, though I think my expectations are pretty 
> reasonable at this time.  I used xf86config to generate a new minimal 
> XF86Config-4 file. The key components of it are:
> 
> Load "dbe"
> Option "omit xfree86-dga"
> Load "type1"
> Load "speedo"
> Driver "vga"
> Driver "vesa"

OK, I guess I'd expected "vga" to work, but you should be able to
do rather better than that.

"vesa" won't work. It relies on call-backs to the BIOS to
configure the display, which we don't currently support. I guess
we could, but there's usually a better driver than vesa available
anyhow, so it hasn't been a priority. 

What happens if you use the "ati" driver? 

Are you sure your kernel is build without CONFIG_{DRM,AGP,FP} ?

If so, its hard to understand why it's rebooting. I wander if
anything might be gleaned from strace'ing what happens when you
start X.

Ian




-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Keir Fraser]

> OK, I guess I'd expected "vga" to work, but you should be able to
> do rather better than that.
> 
> "vesa" won't work. It relies on call-backs to the BIOS to
> configure the display, which we don't currently support. I guess
> we could, but there's usually a better driver than vesa available
> anyhow, so it hasn't been a priority. 
> 
> What happens if you use the "ati" driver? 
> 
> Are you sure your kernel is build without CONFIG_{DRM,AGP,FP} ?
> 
> If so, its hard to understand why it's rebooting. I wander if
> anything might be gleaned from strace'ing what happens when you
> start X.

To get the oops message you will need to connect a null modem serial
cable and grab a serial dump when booting with 'console=ttyS0'.

 -- Keir


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Ian Pratt]

> I'm still having no luck with this, though I think my expectations are pretty 
> reasonable at this time.  I used xf86config to generate a new minimal 
> XF86Config-4 file. The key components of it are:

Paul, 

Please can you give the latest bk repo a go. I fixed a nasty bug
to do with mapping /dev/mem on 2.6.8.1 that must have been
breaking many X drivers and caused memory corruption in the
kernel.

I'd expect the "ati" driver to now work for you -- I've tested it
on a Rage XL card and it works for me.

Cheers,
Ian




-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: reboot weirdness - problem identified

[Paul Dorman]

You guys are great! Ian, that did the trick. Thanks very much for tracking 
that one down. I owe you a pint! :o)

Cheers,
Paul

On Thursday 23 September 2004 05:52 am, Ian Pratt wrote:
> > I'm still having no luck with this, though I think my expectations are
> > pretty reasonable at this time.  I used xf86config to generate a new
> > minimal XF86Config-4 file. The key components of it are:
>
> Paul,
>
> Please can you give the latest bk repo a go. I fixed a nasty bug
> to do with mapping /dev/mem on 2.6.8.1 that must have been
> breaking many X drivers and caused memory corruption in the
> kernel.
>
> I'd expect the "ati" driver to now work for you -- I've tested it
> on a Rage XL card and it works for me.
>
> Cheers,
> Ian
>
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
> Project Admins to receive an Apple iPod Mini FREE for your judgement on
> who ports your project to Linux PPC the best. Sponsored by IBM.
> Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php