Intel D945GNT crashes with AGP enabled

Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi guys,

I have this problem for quite some time now, but I never really got
around to figure out what it is. I can successfully boot this machine
and get X11 up and running, but when I shutdown my machine it crashes
when closing X11. The system is x86_64 based and looks like this:

0000:00:00.0 Host bridge: Intel Corporation 945G/P Memory Controller Hub (rev 02)
0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)
0000:00:1b.0 Audio device: Intel Corporation 82801G (ICH7 Family) High Definition Audio Controller (rev 01)
0000:00:1c.0 PCI bridge: Intel Corporation 82801G (ICH7 Family) PCI Express Port 1 (rev 01)
0000:00:1c.2 PCI bridge: Intel Corporation 82801G (ICH7 Family) PCI Express Port 3 (rev 01)
0000:00:1c.3 PCI bridge: Intel Corporation 82801G (ICH7 Family) PCI Express Port 4 (rev 01)
0000:00:1d.0 USB Controller: Intel Corporation 82801G (ICH7 Family) USB UHCI #1 (rev 01)
0000:00:1d.1 USB Controller: Intel Corporation 82801G (ICH7 Family) USB UHCI #2 (rev 01)
0000:00:1d.2 USB Controller: Intel Corporation 82801G (ICH7 Family) USB UHCI #3 (rev 01)
0000:00:1d.3 USB Controller: Intel Corporation 82801G (ICH7 Family) USB UHCI #4 (rev 01)
0000:00:1d.7 USB Controller: Intel Corporation 82801G (ICH7 Family) USB2 EHCI Controller (rev 01)
0000:00:1e.0 PCI bridge: Intel Corporation 82801 PCI Bridge (rev e1)
0000:00:1f.0 ISA bridge: Intel Corporation 82801GB/GR (ICH7 Family) LPC Interface Bridge (rev 01)
0000:00:1f.1 IDE interface: Intel Corporation 82801G (ICH7 Family) IDE Controller (rev 01)
0000:00:1f.2 IDE interface: Intel Corporation 82801GB/GR/GH (ICH7 Family) Serial ATA Storage Controllers cc=IDE (rev 01)
0000:00:1f.3 SMBus: Intel Corporation 82801G (ICH7 Family) SMBus Controller (rev 01)
0000:04:08.0 Ethernet controller: Intel Corporation 82801G (ICH7 Family) LAN Controller (rev 01)

The problematic part is the Intel AGP module (intel_agp), because if I
don't compile it the system works fine. There is an oops coming, but so
far I wasn't able to get it out. Does anyone have seen this problem
before and have some patches for me to try? Otherwise I need to try to
get this oops message.

Regards

Marcel

Re: Intel D945GNT crashes with AGP enabled

[Alejandro Bonilla]

On Fri, 28 Oct 2005 15:38:35 +0200, Marcel Holtmann wrote
> Hi guys,
> 
> I have this problem for quite some time now, but I never really got
> around to figure out what it is. I can successfully boot this machine
> and get X11 up and running, but when I shutdown my machine it crashes
> when closing X11. The system is x86_64 based and looks like this:

Marcel,

This is the Linux kernel development ML, if you have a problem, you should
really try hard to see what could be wrong.

1. oops is of good help.
2. go into a tty (ctrl+alt+f1) and try moving between runlevels and see if you
notice anything. Additionally see if this only occurs when only shuting down,
or rebooting or if also when moving from runlevels.
3. Could this also be a distro problem associated with the intel_agp module in X?

Ask yourself questions, do some googling and once you have some outputs then
maybe people can have something for you here.

.Alejandro

Re: Intel D945GNT crashes with AGP enabled

[Dave Jones]

On Fri, Oct 28, 2005 at 03:38:35PM +0200, Marcel Holtmann wrote:
 > The problematic part is the Intel AGP module (intel_agp), because if I
 > don't compile it the system works fine. There is an oops coming, but so
 > far I wasn't able to get it out. Does anyone have seen this problem
 > before and have some patches for me to try? Otherwise I need to try to
 > get this oops message.

You never mentioned what kernel you're running.
If it's a recent -mm, there's an AGP optimisation patch to do less
frequent TLB flushes, which may be worth backing out.

If you're running mainline, I'm puzzled.
It'd be useful to see that oops.

		Dave

Re: Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi Dave,

>  > The problematic part is the Intel AGP module (intel_agp), because if I
>  > don't compile it the system works fine. There is an oops coming, but so
>  > far I wasn't able to get it out. Does anyone have seen this problem
>  > before and have some patches for me to try? Otherwise I need to try to
>  > get this oops message.
> 
> You never mentioned what kernel you're running.
> If it's a recent -mm, there's an AGP optimisation patch to do less
> frequent TLB flushes, which may be worth backing out.
> 
> If you're running mainline, I'm puzzled.

basically I am running mainline, but I also tried your agpgart tree and
both are having problems.

> It'd be useful to see that oops.

I am working on it and actually it is enough to just kill the X process
to crash the system. I saw parts of the oops and it seems that there are
some RCU calls in the backtrace, but this might not help at all. So it
seems that I really need that oops.

Regards

Marcel

Re: Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi Dave,

> >  > The problematic part is the Intel AGP module (intel_agp), because if I
> >  > don't compile it the system works fine. There is an oops coming, but so
> >  > far I wasn't able to get it out. Does anyone have seen this problem
> >  > before and have some patches for me to try? Otherwise I need to try to
> >  > get this oops message.
> > 
> > You never mentioned what kernel you're running.
> > If it's a recent -mm, there's an AGP optimisation patch to do less
> > frequent TLB flushes, which may be worth backing out.
> > 
> > If you're running mainline, I'm puzzled.
> 
> basically I am running mainline, but I also tried your agpgart tree and
> both are having problems.

I also checked a non-SMP and non-preempt kernel. Both make my system
crash when killing the X server.

> > It'd be useful to see that oops.
> 
> I am working on it and actually it is enough to just kill the X process
> to crash the system. I saw parts of the oops and it seems that there are
> some RCU calls in the backtrace, but this might not help at all. So it
> seems that I really need that oops.

So getting the oops is really hard. I can't get it via serial port and
the netconsole (it uses an e100) is also not working. I tried to run
framebuffer, but I can only use vesafb and the resolution is nothing
better than any text console and I wasn't able to change it. So taking a
picture of the oops is also not an option. What other possible ways I
have to finally get the oops out of the system?

And btw why can't I compile the intelfb on x86_64? I use the internal
graphics card (8086:2772) of the D945GNT motherboard.

0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)

The Kconfig file makes it unselectable on x86_64 system.

config FB_INTEL
        tristate "Intel 830M/845G/852GM/855GM/865G support (EXPERIMENTAL)"
        depends on FB && EXPERIMENTAL && PCI && X86 && !X86_64

It seems that the most recent support in this driver is for the 915G and
not for the 945G. Are they so different that we need a complete new
driver or is the !X86_64 a relict from old times?

Regards

Marcel

Re: Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi Dave,

> And btw why can't I compile the intelfb on x86_64? I use the internal
> graphics card (8086:2772) of the D945GNT motherboard.
> 
> 0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)
> 
> The Kconfig file makes it unselectable on x86_64 system.
> 
> config FB_INTEL
>         tristate "Intel 830M/845G/852GM/855GM/865G support (EXPERIMENTAL)"
>         depends on FB && EXPERIMENTAL && PCI && X86 && !X86_64
> 
> It seems that the most recent support in this driver is for the 915G and
> not for the 945G. Are they so different that we need a complete new
> driver or is the !X86_64 a relict from old times?

it seems that the first problem is:

arch/x86_64/kernel/built-in.o: In function `pci_iommu_init':
pci-gart.c:(.init.text+0x8f0c): undefined reference to `agp_amd64_init'
pci-gart.c:(.init.text+0x8f1a): undefined reference to `agp_bridge'
pci-gart.c:(.init.text+0x8f1f): undefined reference to `agp_copy_info'
make: *** [.tmp_vmlinux1] Error 1

Is this fixable or will the intelfb never work on x86_64 system?

Regards

Marcel

Re: Intel D945GNT crashes with AGP enabled

[Alan Hourihane]

On Sat, 2005-10-29 at 14:27 +0200, Marcel Holtmann wrote:
> Hi Dave,
> 
> > And btw why can't I compile the intelfb on x86_64? I use the internal
> > graphics card (8086:2772) of the D945GNT motherboard.
> > 
> > 0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)
> > 
> > The Kconfig file makes it unselectable on x86_64 system.
> > 
> > config FB_INTEL
> >         tristate "Intel 830M/845G/852GM/855GM/865G support (EXPERIMENTAL)"
> >         depends on FB && EXPERIMENTAL && PCI && X86 && !X86_64
> > 
> > It seems that the most recent support in this driver is for the 915G and
> > not for the 945G. Are they so different that we need a complete new
> > driver or is the !X86_64 a relict from old times?
> 
> it seems that the first problem is:
> 
> arch/x86_64/kernel/built-in.o: In function `pci_iommu_init':
> pci-gart.c:(.init.text+0x8f0c): undefined reference to `agp_amd64_init'
> pci-gart.c:(.init.text+0x8f1a): undefined reference to `agp_bridge'
> pci-gart.c:(.init.text+0x8f1f): undefined reference to `agp_copy_info'
> make: *** [.tmp_vmlinux1] Error 1
> 
> Is this fixable or will the intelfb never work on x86_64 system?

You are barking up the wrong tree with this. Read the tristate line.

It doesn't mention 915 or 945 support. Intelfb only supports upto the
865G anyway.

Alan.

Re: Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi Alan,

> > > And btw why can't I compile the intelfb on x86_64? I use the internal
> > > graphics card (8086:2772) of the D945GNT motherboard.
> > > 
> > > 0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)
> > > 
> > > The Kconfig file makes it unselectable on x86_64 system.
> > > 
> > > config FB_INTEL
> > >         tristate "Intel 830M/845G/852GM/855GM/865G support (EXPERIMENTAL)"
> > >         depends on FB && EXPERIMENTAL && PCI && X86 && !X86_64
> > > 
> > > It seems that the most recent support in this driver is for the 915G and
> > > not for the 945G. Are they so different that we need a complete new
> > > driver or is the !X86_64 a relict from old times?
> > 
> > it seems that the first problem is:
> > 
> > arch/x86_64/kernel/built-in.o: In function `pci_iommu_init':
> > pci-gart.c:(.init.text+0x8f0c): undefined reference to `agp_amd64_init'
> > pci-gart.c:(.init.text+0x8f1a): undefined reference to `agp_bridge'
> > pci-gart.c:(.init.text+0x8f1f): undefined reference to `agp_copy_info'
> > make: *** [.tmp_vmlinux1] Error 1
> > 
> > Is this fixable or will the intelfb never work on x86_64 system?
> 
> You are barking up the wrong tree with this. Read the tristate line.
> 
> It doesn't mention 915 or 945 support. Intelfb only supports upto the
> 865G anyway.

I read that line and enabling the compilation of the intelfb on x86_64
was only a simple try to see if it compiles. It doesn't and basically I
have no idea how to fix it.

The 915G (8086:2582) is supported by the driver in the latest vanilla
kernel. Look at intelfb.h file and then you will see that the comment
line in Kconfig is outdated. I also found a patch to support the 915GM
(8086:2592) and maybe a similar patch will make the 945 work with the
intelfb driver.

Regards

Marcel

Re: Intel D945GNT crashes with AGP enabled

[Alan Hourihane]

On Sat, 2005-10-29 at 15:55 +0200, Marcel Holtmann wrote:
> Hi Alan,
> 
> > > > And btw why can't I compile the intelfb on x86_64? I use the internal
> > > > graphics card (8086:2772) of the D945GNT motherboard.
> > > > 
> > > > 0000:00:02.0 VGA compatible controller: Intel Corporation 945G Integrated Graphics Controller (rev 02)
> > > > 
> > > > The Kconfig file makes it unselectable on x86_64 system.
> > > > 
> > > > config FB_INTEL
> > > >         tristate "Intel 830M/845G/852GM/855GM/865G support (EXPERIMENTAL)"
> > > >         depends on FB && EXPERIMENTAL && PCI && X86 && !X86_64
> > > > 
> > > > It seems that the most recent support in this driver is for the 915G and
> > > > not for the 945G. Are they so different that we need a complete new
> > > > driver or is the !X86_64 a relict from old times?
> > > 
> > > it seems that the first problem is:
> > > 
> > > arch/x86_64/kernel/built-in.o: In function `pci_iommu_init':
> > > pci-gart.c:(.init.text+0x8f0c): undefined reference to `agp_amd64_init'
> > > pci-gart.c:(.init.text+0x8f1a): undefined reference to `agp_bridge'
> > > pci-gart.c:(.init.text+0x8f1f): undefined reference to `agp_copy_info'
> > > make: *** [.tmp_vmlinux1] Error 1
> > > 
> > > Is this fixable or will the intelfb never work on x86_64 system?
> > 
> > You are barking up the wrong tree with this. Read the tristate line.
> > 
> > It doesn't mention 915 or 945 support. Intelfb only supports upto the
> > 865G anyway.
> 
> I read that line and enabling the compilation of the intelfb on x86_64
> was only a simple try to see if it compiles. It doesn't and basically I
> have no idea how to fix it.
> 
> The 915G (8086:2582) is supported by the driver in the latest vanilla
> kernel. Look at intelfb.h file and then you will see that the comment
> line in Kconfig is outdated. I also found a patch to support the 915GM
> (8086:2592) and maybe a similar patch will make the 945 work with the
> intelfb driver.

Doing a quick google shows that support is only partial for 915G. And
the 915GM patch would be easily done for the 945G.

But I'm not sure how partial that support is having not tried it, but I
can see that there are definate problems here.

Alan.

Re: Intel D945GNT crashes with AGP enabled

[Marcel Holtmann]

Hi Alan,

> > > You are barking up the wrong tree with this. Read the tristate line.
> > > 
> > > It doesn't mention 915 or 945 support. Intelfb only supports upto the
> > > 865G anyway.
> > 
> > I read that line and enabling the compilation of the intelfb on x86_64
> > was only a simple try to see if it compiles. It doesn't and basically I
> > have no idea how to fix it.
> > 
> > The 915G (8086:2582) is supported by the driver in the latest vanilla
> > kernel. Look at intelfb.h file and then you will see that the comment
> > line in Kconfig is outdated. I also found a patch to support the 915GM
> > (8086:2592) and maybe a similar patch will make the 945 work with the
> > intelfb driver.
> 
> Doing a quick google shows that support is only partial for 915G. And
> the 915GM patch would be easily done for the 945G.
> 
> But I'm not sure how partial that support is having not tried it, but I
> can see that there are definate problems here.

I can do the patch for the 945G by myself, but the current problem is
that the intelfb driver not even compiles on EM64T. Otherwise I would
give it a try and see if it helps me.

Regards

Marcel