Re: Radeon 7500 w/ DRI locking on restart of X

I have precisely the same symptoms as what Glenn listed.

I have now tried the forcepcimode option on the laptop and
unfortunately experienced what appears to be the opposite
effect to what Glenn noted.  I get a black screen with lots
of disk activity.  I can log in remotely (at least that
seemed to work every time) and a top listing shows X taking
up more and more CPU cycles.  It seems to be scribbling to
the disk making temporary files as well because eventually
I couldn't edit the XF86Config file any longer -- I was
getting a "no space left on device" error.  I ended up 
taking out all options short of disabling DRI altogether
and eventually discovered the only way I could run with 
DRI switches on and forcepcimode set to true was to unload
the agp module from kernel space.  This got it started just
fine but there was no DRI.  It seems that X is getting
confused and somehow PCI and AGP modes are duking it out...

Here is a listing of the X modules from ports:

XFree86-4.3.0,1
XFree86-FontServer-4.3.0
XFree86-Server-4.3.0_5
XFree86-clients-4.3.0_1
XFree86-documents-4.3.0
XFree86-font100dpi-4.3.0
XFree86-font75dpi-4.3.0
XFree86-fontCyrillic-4.3.0
XFree86-fontDefaultBitmaps-4.3.0
XFree86-fontEncodings-4.3.0
XFree86-fontScalable-4.3.0
XFree86-libraries-4.3.0_3
Xaw3d-1.5
Xft-2.1_8

This is what I see from 'dmesg | grep drm' :

drm0: <ATI Radeon LW Mobility 7 (AGP)> port 0xcc00-0xccff mem 0xfcff0000-0xfcffffff,0xe0000000-0xe7ffffff irq 11 at device 0.0 on pci1
info: [drm] AGP at 0xe8000000 64MB
info: [drm] Initialized radeon 1.8.0 20020828 on minor 0

Here is the output of 'head /var/log/XFree86.0.log' :


XFree86 Version 4.3.0 (DRI trunk)
Release Date: 27 February 2003
X Protocol Version 11, Revision 0, Release 6.6
Build Operating System: FreeBSD 4.8-RELEASE i386 [ELF] 
Build Date: 09 August 2003
        Before reporting problems, check http://www.XFree86.Org/
        to make sure that you have the latest version.
Module Loader present
Markers: (--) probed, (**) from config file, (==) default setting,

The above was built on 08/03/03 in the evening.

Here is 'uname -a':

FreeBSD NitroPhys.welchsmnet.net 4.8-RELEASE FreeBSD 4.8-RELEASE #1: Thu May 29 11:24:46 CDT 2003     welchsm_at_NitroPhys.welchsmnet.net:/usr/src/sys/compile/NITROPHYS  i386

The kernel is compiled with SSE support.

I have tried turning off all options in the config file and
different AGP modes.  I always get the same results.

My 5.1-RELEASE install is just using the X off the install
disc with the kernel module that shipped with it.  Same exact
issues.

I believe I saw a commit to DRI CVS a while back that was 
supposed to address this very issue (comment said something 
about locks on logout with GDM) but it didn't change anything
for me.

Let me know what else you'd like to have in the way of info.

                                                  Sean

-------Original Message-------
From: Glenn Johnson <gjohnson_at_srrc.ars.usda.gov>
Sent: 08/26/03 03:55 PM
To: Eric Anholt <eta_at_lclark.edu>
Subject: Re: Radeon 7500 w/ DRI locking on restart of X

> 
> On Tue, Aug 26, 2003 at 01:27:11PM -0700, Eric Anholt wrote:

> On Tue, 2003-08-26 at 18:05, Vulpes Velox wrote:
>
> > I had similar problem with a 7200 and OGL. I solved the problem by
> > turning off some of the options in the X config.
> >
> > On Tue, 26 Aug 2003 12:21:56 -0500 (GMT) Sean Welch
> > <welchsm_at_earthlink.net> wrote:
> >
> > > Is anyone else seeing this issue?  I'm running into it on desktop
> > > boxes and a laptop running 4.8-RELEASE with up to date ports
> > > collections and various versions of DRI installed over a ports
> > > version of X.  I'm also seeing this under 5.1-RELEASE on the
> > > laptop.
> > >
> > > Everything works perfectly unless/until I restart the X server.
> > > This appears to be initiated automatically when running GDM -- ie,
> > > GDM starts, you log in using that X session, you log out and the
> > > session stops, GDM starts X again and displays the login screen.
> > >
> > > This seems to happen a bit more than 1/3 of the times I try it
> > > (intentionally or not).  It isn't much of a problem on the laptop
> > > as I'm the only user and tend to turn the machine off when I log
> > > out but it is causing all sorts of issues on the desktops because
> > > they are intended to be used as multi-user (serially and also
> > > simultaneously) systems.
> > >
> > > Any ideas?  The instability goes away completely with DRI
> > > disabled, but part of the use of these desktops is in the
> > > accelerated OpenGL rendering...
> > >
> > >                                                   Sean
>
> (CC'ed to -current since it's not -stable specific)
>
> This is an example of why people need to send PRs and not just emails.
> I realize now I've seen emails on this before, but I had forgotten
> about them because they seemed isolated and I didn't have them piling
> up in my assigned prs list (things pile up in my mailboxes easily, and
> I don't go back and check very often).

Speaking for myself, I was not sure whether the problem was a BIOS
setting problem, a graphics card problem, a FreeBSD problem, an XFree86
problem, or a configuration problem on my part.  Until today, I thought
I was the only one with seeing this.  I discovered the "ForcePCIMode"
option as a potential workaround only two days ago and I wanted to
confirm that it did indeed make the problem go away before filing a PR
and sending an e-mail directly to you about the issue.  In fact, I was
planning on doing that this evening after a few more tests.

> I've tried to reproduce this with a radeon, by doing startx, C-A-B
> to kill the server, then startx again.  The second time, the screen
> displays for a brief moment then goes black.  The system isn't hung,
> and I can exit using C-A-B again.  Is this what everyone else sees?

Sometimes.  At times the Xserver will not start and I can switch to a
console and login.  Any further attempts to restart X will lock the
machine up.  Other times, the screen just goes black and the machine
is locked up.  I can not switch to a console nor ssh in via another
machine.  In that case, I have to hit the reset switch.  Still other
times, the gdm login screen will appear and look normal, except the
cursor is not blinking and the keyboard and mouse do not respond.  I did
not try to ssh into the machine in that state but I would guess that it
would fail.

> Everyone that's experiencing this and is using the DRI, what version
> of the radeon DRM is loaded? (dmesg | grep drm)

I will have to get the dmesg output for you when I get home but it is
the latest version in -current, with -current being up to date as of Aug
25, 2003, about 9:00 PM CDT.

> Is anyone experiencing this without the DRI loaded?

Not me; when I disable DRI the lock ups do not occur.

> The ForcePCIMode workaround is interesting, I'll take a look at what
> could be going on there.

I have had that option in my config file for about two days and have
tested by logging in and out repeatedly.  So far, I have not had a lock
up with that option enabled.

-- 
Glenn Johnson
USDA, ARS, SRRC   	   	   	 Phone: (504) 286-4252
New Orleans, LA 70124   	   	e-mail: gjohnson_at_srrc.ars.usda.gov
>