Common radeon/radeonhd X restart lockup problem

[Paul Knowles]

Hello, 

I've a lockup problem common to both the radeon and radeonhd
drivers, and it is somewhat weird to diagnose.  I have been
testing on three different machines, only one manifests the
problem:

1: ATI Technologies Inc Radeon X1650 Pro (rev 9e) (ChipID =
   0x71c1) connected to a Asus P3E5 mobo via PCIe 2.0.
   x86_64, FC9

2: ATI Technologies Inc Radeon X1650 Pro (rev 9e) (ChipID =
   0x71c1) connected to an Intel D856GLC mobo, AGP, ix86 FC9

3: ATI Technologies Inc RV505 CE [Radeon X1550 64-bit]
   (ChipID = 0x715f) on PCIe 2.0, asus mobo, Athlon(tm) 64
   processor x86_64 FC9

All three machines have both
   xorg-x11-drv-ati-6.8.0-19.fc9.x86_64.rpm
   xorg-x11-drv-radeonhd-1.2.1-3.7.20080724git.fc9.x86_64.rpm
installed from fedora updates-testing

runlevel is 3 by default: X is started and stopped as needed.

The ...drv-ati, i.e., radeon, driver is in use since it
provides Xv overlay support with good results in mplayer,
and good acceleration in 3D.  On machines 2 and 3, I can
startx, stop and restart without problems.  I can even
startx on one VT, switch out to the login VT, and startx
again and have a second x session running (2nd session
without DRI, of course).

On machine 1, I can startx, VT switch in and out of the X
session, but if I stop the X session, the next startx causes
a hard lockup. This happens for both radeon and radeonhd
drivers, with both EXA and XAA acceleration modes.  VT
switching out and starting a second X session works, however
exiting either of the running X session also causes a hard
lockup.

xorg driver section:

Section "Device"
        Identifier  "radeon_exa"
#        Driver      "radeonhd"
        Driver      "radeon"
        VendorName  "ATI Technologies Inc"
        Option      "ModeDebug" "true"
        Option      "backingstore" "true"
        Option      "MergedXinerama" "false"
        Option      "MergedFB" "false"
# attacking the flickering problems
        Option      "DRI" "on"
#        Option      "AccelMethod" "XAA"
        Option      "AccelMethod" "EXA"
        Option      "EnablePageFlip" "1"
        Option      "ColorTiling" "1"
        Option      "DisplayPriority" "high"
EndSection

Following the lockup and reboot, the Xorg.?.log shows no
indication of problems.

This is for me very odd since machines 1 and 3 are the same
x86_64 software, but 1 locks up and 3 doesn't.  Machines 1
and 2 are the same ATI chip, but different software arch and
different bus: 1 locks up and 2 doesn't.  I borrowed a
FireGL 3350 and tried with that in machine 1, with identical
results to the crashes above.  With the vesa driver, all
machines work the same way, X starts and stops as it should,
even with multiple sessions on many VTs.

Since the problem happens for both radeon and radeonhd, at
startup on the second invocation, I suspect something in the
initialization code common to both drivers.

Any suggestions?

.paul.