Re: [gentoo-user] Weird GNOME problem

[Holly Bostick]

Michael Sullivan schreef:
> On Sat, December 31, 2005 4:44 am, Rumen Yotov wrote:
>
>> On (31/12/05 01:49), michael@××××××××××××.com wrote:
>>
>>> Earlier tonight I changed the setting in /etc/conf.d/rc for the
>>> RC_DEVICE_TARBALL from "yes" to "no" and rebooted.  Now I can't
>>> get into GNOME.  For my personal account after I enter my
>>> username/password the screen clears and I see the default
>>> background color and my mouse pointer, then nothing else.  I can
>>> move the mouse pointer, but there's nothing the click on and the
>>> context menu doesn't come up when I right-click.  When I tried to
>>> log in as root from the welcome screen I see the "GNOME Starting"
>>> box with the Gentoo logo, but it never goes on from that.  To get
>>> out of both my personal account and the root account I had to
>>> Alt+Cntrl+Backspace.  I got a Failsafe Terminal and looked at the
>>> log files for gdm and saw this:
>>>
>>> (EE) GARTInit: Unable to open /dev/agpgart (No such file or
>>> directory)
>>>
>>> There are five log files in /var/log/gdm; they all say roughly
>>> the same thing, and their timestamps are consistent with the
>>> times I tried to log in and got stuck.  I can't even find out why
>>> it's looking for this file or how to make it stop looking for it.
>>> Can anyone help me out here? -- gentoo-user@g.o mailing
>>> list
>>>
>>
>> Hi, Are you using a Nvidia video card? Reverting RC_DEVICE_TARBALL
>> to "yes" should solve the problem but it's not a solution. udev
>> isn't creating the /dev/agpgart file/link - check if it's there and
>>  with what perms. Rumen
>>
>
>
> /dev/agpgart does not exist.  As far as I know I am not using a
> Nvidia card.  If I was I think I would have a /dev/nvidia as well as
> several other similar /dev files starting with nvid, but I don't...
>

"As far as you know"? Do you not know what video card you have in your
box? This is not really a wise policy in general (not knowing what your
hardware is), but especially not when running Linux.

But anyway, here's the deal.

AGPGART is the device representing your AGP bus, which bus (slot) most
likely holds your video card (unless you have an onboard video chip, a
PCI video card, or use PCI-E, which you might, but most likely do not,
as the vast majority of systems in use today use an AGP video card).

AGPGART is a motherboard resource, which is part of the kernel. What may
have happened is that you upgraded your kernel at some point and did not
enable agpgart in the kernel config, and/or did not enable specific
support for your motherboard, but this was not noticed because the
previously-existing device file was in the tarball you were using.
However, when you stopped using the tarball, because the modules were
not present in the kernel, a new device could not be created (because no
module/driver was found to create the device necessary). This is, of
course, all speculation, but I can't offhand think of any other reason
that /dev/agpgart would not be created, other than that the module isn't
loaded, and the most likely reason that that would happen is because
there is no support in the kernel enabled. However, this could also
happen because the module isn't explicitly loaded because either
coldplug is not installed, and/or the module--if compiled as a module--
needs to be in /etc/modules.autoload.d/kernel-2.* and is not there
present, and/or the module-- if compiled directly into the kernel is not
loading because the kernel itself does not have support for loading
modules enabled. But these are all somewhat less likely than agpgart
support not being enabled in the first place (which is itself fairly
unlikely, but since you are not getting the device being created, there
must be *some* reason that that isn't happening).

What Rumen is talking about is a specific quirk of nVidia cards using
the proprietary nVidia drivers, that the device must be created in
local.start because the device is for some reason not created
automatically (I don't have an nVidia card, so I don't know the
details). There is also a specific quirk of ATI cards using the
proprietary fglrx drivers, that agpgart must be compiled into the kernel
as a module, but your xorg.conf must be specifically set to either use
the kernel module or the module included with the fglrx drivers; the two
are incompatible and the fglrx drivers may or may not work if you use
the wrong one.

The reason you can get into a failsafe GNOME session is because that
uses the 'vesa' video driver, which is practically guaranteed to provide
a basic display (it's basically like Windows Safe Mode, doesn't run
anything except what you absolutely need to have enough of a session
that you can fix whatever's wrong). However, a full GNOME session
requires the driver specified in your xorg.conf to load, which it most
likely cannot, because your AGP bus is not loaded, so X is unable to
communicate with your video card (which is on the AGP bus, which is not
accessible, because the device file does not exist).

So what we want is to cause this device file to be created, which would
mean that we have to know somewhat more about

1) what kernel you are using

2) how it is configured

3) what specific hardware exists in the system (which impacts the
configuration of the kernel).

For the time being, I would suggest re-enabling the tarball in
/etc/conf.d/rc, so that you can have a fully-working system while you
determine the status of Device Drivers=> Character Devices=>
/dev/agpgart (AGP support) in your current kernel config, and the nature
of your motherboard and video card, so that you know what the correct
settings for your system actually need to be.

HTH,
Holly
-- 
gentoo-user@g.o mailing list