Hi,

 on one of my boxes, I've got a problem with gdm and kernels newer
than 2.6.24 (tested on 2.6.24.2, 2.6.24.4).  If I try to restart or
shut down from gdm, the window disappears but the X background remains
and the box stays in runlevel 5 until I switch to a tty and shut it
down (as root) or give it a 3-fingered salute to reboot.

 The same with 2.6.25-rc6-git8.

 The only oddity in the logs is a large block of
Mar 24 13:49:29 bluesbreaker gdm[2554]: Handling user message:
'GET_CONFIG greeter/SetPosition :0'
Mar 24 13:49:29 bluesbreaker gdmlogin[2995]:   Got response: 'OK
false'
Mar 24 13:49:29 bluesbreaker gdmlogin[2995]: Sending command:
'CLOSE'
Mar 24 13:49:29 bluesbreaker gdm[2554]: Handling user message:
'CLOSE'
Mar 24 13:49:29 bluesbreaker gdm[2562]: gdm_slave_wait_for_login: In
loop
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login:
end verify for ''
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login: No
login/Bad login
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login: In
loop
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login:
end verify for ''
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login: No
login/Bad login
Mar 24 13:49:35 bluesbreaker gdm[2562]: gdm_slave_wait_for_login: In
loop
... about 165 repeats of these 3 lines, it seems to stop the
messages on its own, then I ran shutdown and got the normal

Mar 24 13:50:02 bluesbreaker shutdown[3000]: shutting down for
system halt
Mar 24 13:50:02 bluesbreaker init: Switching to runlevel: 0
Mar 24 13:50:08 bluesbreaker logger: /etc/rc.d/init.d/rc called with
arg 0
Mar 24 13:50:08 bluesbreaker logger: /etc/rc.d/rc0.d/K01cups called
with arg stop
 and so forth.

 Yes, I've got a lot of debugging in there, I used this box to find
out why gdm-2.20 gave me a messy shutdown, and left it in after I
reverted to gdm-2.18.

 This only happens with 64-bit kernels, I've been using it as i686
without any issues.

 While trying to debug this, I reverted the two drm patches from
2.6.24.1 (in 2.6.24.4) and normal behaviour was restored.  I forgot
to change the extraversion, which meant my modules were overwritten.
When I later went back to the "bad 2.6.24.4" it shut down correctly.
Examination showed i2c_viapro was the key - if I rmmod it in
2.6.24.x, gdm works correctly.  For the record, my 32-bit config is
very different, but does include i2c_viapro as a module.

 So, I started to think it might be an i2c problem.  But with
2.6.25-rc6-git8 I can't get gdm to work correctly, neither by
reverting the patches, nor with rmmod, so for the moment I'm in the
dark.  My notes say that I managed to fix it one time on -rc6-git8,
by 'rmmod i2c_viapro' before logging out, but all subsequent attempts
to repeat this have failed.

 FWIW, the via ISA bridge is a 1106:3227 (VIA_8237, also
described as VT8237R in i2c-viapro).

 Going back to 2.6.24.4 and reverting only the following patch,
normal behaviour is restored:

diff --git a/drivers/char/drm/drm_vm.c b/drivers/char/drm/drm_vm.c
index e8d50af..ef5e6b1 100644
--- a/drivers/char/drm/drm_vm.c
+++ b/drivers/char/drm/drm_vm.c
@@ -506,6 +506,7 @@ static int drm_mmap_dma(struct file *filp,
struct vm_area_struct *vma)
        vma->vm_ops = &drm_vm_dma_ops;
 
        vma->vm_flags |= VM_RESERVED;   /* Don't swap */
+       vma->vm_flags |= VM_DONTEXPAND;
 
        vma->vm_file = filp;    /* Needed for drm_vm_open() */
        drm_vm_open_locked(vma);
@@ -655,6 +656,7 @@ static int drm_mmap_locked(struct file *filp,
struct vm_area_struct *vma)
                return -EINVAL; /* This should never happen. */
        }
        vma->vm_flags |= VM_RESERVED;   /* Don't swap */
+       vma->vm_flags |= VM_DONTEXPAND;
 
        vma->vm_file = filp;    /* Needed for drm_vm_open() */
        drm_vm_open_locked(vma);

 I haven't been following 2.6.25 closely, but I assume that what got
applied to 2.6.24-stable was only part of the change to 2.6.25, and
perhaps some other part of it is affecting me ?

 .config.gz for 2.6.25-rc6-git8 is attached.

Ken
-- 
das eine Mal als Tragödie, das andere Mal als Farce