[ltp] Re: Success with fglrx 8.28.8?

[TNKS]

I found out that this problem is pretty wide-spread after digging into the
unofficial ATI Bugzilla (I forgot about that page).  Here's one of the
links, but there are other bugs on there that I think are duplicates of the
same issue.

        http://ati.cchtml.com/show_bug.cgi?id=239

After hacking around, though, I figured somethings out that aren't yet on
the bugzilla page (I'll probably post there myself soon).


CHARACTERIZING THE PROBLEM
--------------------------

The X VT (VT7 for most people) hangs, but you /can/ still switch to other
VT's.  The problem is that the screen is blank.  But, you can blind type in
the text text VTs. 

The reason I say that the X VT is hung is because if I try to use acpid to
ACPI S3 sleep the system using Fn-F4 while hung in the X VT, nothing
happens.  But the moment I Ctrl-Alt-F1, the system proceeds to suspend. 
It's as though the request to sleep was stuck behind some process that
won't finish until you get out of the X VT.

Even more interesting, upon resuming from the S3 sleep, everything is back
to normal.  My display is back.

I thought that maybe the Suspend2 hibernate script was doing something fancy
to get my display back.  So I did a hibernate with the -n switch.  This
does all the script activity that surrounds sleep cycle, but doesn't
actually do the sleep itself.  The -n switch didn't bring back my display. 
So I'm concluding that it's the ACPI S3 sleep itself that fixes the
display.


ISOLATING THE PROBLEM
---------------------

Another note, I tried the latest 8.28.8 with an identical kernel to mine (I
use the latest Debian-patched 2.6.17-7 source package in Sid), but without
any framebuffer functionality compiled in.  I got the same problem, so I
don' think the problem is related to the frame buffer.  Furthermore, I get
the problem with and without the fglrx kernel module loaded, and with and
without DRI/GLX enabled in my xorg.conf

Finally, there is one way I've found to get rid of the problem -- use the
Debian binary linux-image-2.6.16-2-686 kernel package.  The interesting
thing I found, though is that the problem is there on the Debian binary
linux-image-2.6.17-2-686 kernel package.  

In summary, it seems that fglrx 8.28.8 introduced a bug that wasn't there in
8.27.10 that shows itself in a 2.6.17 kernel, but not in the 2.6.16 kernel. 
This problem seems to be independent of whether DRI/GLX is enabled, whether
the kernel provides a framebuffer, or whether the fglrx kernel module is
loaded.


I'm happy to continue discussion on this thread, but probably only if it's
something beyond what's already posted on their Bugzilla.


Thanks for the responses, though.  I'll probably put some revision of this
E-mail on the ATI Bugzilla now.

- Sukant