I found that the X server does restore the console - sometimes.

I do not see any pattern in it. It just sometimes breaks. However, once broken
the console does not come back.

Not sure if my problem is identical but I am seeing a very similar regression in
master.  Exiting the server does not return to the console command prompt but
instead gives a blank screen with blinking cursor that does not respond to
Ctrl+Alt+F1.

What's worse is that running a second server does not work.  Some teardown is
not being performed properly.  I have attached both log files, the first server
(which runs but doesn't exit properly as described above) and the second server
(which won't run) as well as xorg.conf.

Using git-bisect I narrowed it down to the following commit: a6381e69....:
xfree86 DDX: Delete DDX screens in ddxGiveUp().  I can confirm that this change
introduces the bug; removing the three lines from xf86Init.c solves the problem
for me.

My hardware an is IBM ThinkPad T60p with a recent Radeon card running FC5. 
Video driver is VESA, standard keyboard & mouse drivers.

Created attachment 7992 [details]
Log of first xserver run

Created attachment 7993 [details]
Log of second xserver run

Created attachment 7994 [details]
configuration file

(In reply to comment #2)
> Not sure if my problem is identical 

I rather doubt it, as the regression you found shouldn't affect VT switching. I
suspect the original problem is driver specific, so please attach a full log
file Michal.


> What's worse is that running a second server does not work.  Some teardown is
> not being performed properly.

Ironically, the purpose of the change that caused the regression was to enable
teardown that wasn't done before... looks like that indeed does more than
expected, putting the graphics card into a bad state. I may look into this
later, but I just reverted the change for now.

Created attachment 8004 [details]
log of running X server when switching is already broken

(In reply to comment #0)
> I noticed that X -configure prints the card type on screen, and so does the X
> server on startup. Is this done in graphics?

No, sounds like the video BIOS POST output.

> The same radeon driver with a 7.1 X server worked.

Did you rebuild the driver against the server in both cases? As you seem to be
running the driver from git, can you check whether this also happens with the
latest release?

I recompiled the driver for the server afaict.

I do not use the 7.1 xorg release because OpenGL does not work on Radeon with that.

But I might be able to recompile the old driver and see if it fails to restore
console as well.

I did not try the old driver yet but I was unable to reproduce the problem with
AIGLX disabled.

With ati 6.6.3 it is harder to reproduce. Perhaps it happens only when the
server terminates.

(In reply to comment #11)
> With ati 6.6.3 it is harder to reproduce.

Can you try isolating the change that made things worse with git-bisect?

(In reply to comment #9)
> I do not use the 7.1 xorg release because OpenGL does not work on Radeon with
that.

What's the problem with that?

 (In reply to comment #12)
> (In reply to comment #11)
> > With ati 6.6.3 it is harder to reproduce.
> 
> Can you try isolating the change that made things worse with git-bisect?

Unfortunately I probably won't be able to do that now.

> 
> (In reply to comment #9)
> > I do not use the 7.1 xorg release because OpenGL does not work on Radeon with
> that.
> 
> What's the problem with that?

The normal stuff. The card locks up and the system has to be rebooted to get
anything on the screen again.

(In reply to comment #12)
> (In reply to comment #11)
> > With ati 6.6.3 it is harder to reproduce.
> 
> Can you try isolating the change that made things worse with git-bisect?
> 

Was the patch to fix the console on exit already checked in? I tried with some X
server after Dec 20, and I still do get broken console after server exit, even
with the release driver.

Without fixing that it would be quite hard to find the version that also breaks
the conslole while running.

(In reply to comment #14)
> 
> Was the patch to fix the console on exit already checked in? 

Which patch are you referring to?

(In reply to comment #15)
> (In reply to comment #14)
> > 
> > Was the patch to fix the console on exit already checked in? 
> 
> Which patch are you referring to?

There is not really a patch - the change explained in comment #2.

I fixed that regression in
http://gitweb.freedesktop.org/?p=xorg/xserver.git;a=commit;h=729fca33a417ae3dfb180caf0ea8946ef9eee1df
almost a month ago.

ok, so this is probably something else. I can see the change reverted in the
tree I have here.

I will try to find a configuration that does restore the console correctly, even
on exit.

iirc it worked with the old X server, even with recent ati driver.

Now with the latest ati driver I can no longer reproduce the problem.

There was some fix related to saved state so I guess that's it.

(In reply to comment #19)
> Now with the latest ati driver I can no longer reproduce the problem.
> 
> There was some fix related to saved state so I guess that's it.
If you are refering to 9506f7015a1e442f4ca0bd3bfae555ec7e8a5f37 it shouldn't
make a difference, as this should only affect the register values in use while X
is active, but the driver should restore all values as read initially when
switching to a console, which this commit does not change.