X server on Matrox g550 is all black (no output)

[linas]

No video output (all black screen) on matrox g550. Evidence below, but here's a wild guess: this card has a dual-dvii output, and I'm guessing that the mga driver is configured to send the video signal to a non-existent vga port (i.e. older models of g550 had 1 vga, 1 dvi, but this one has 2 dvi). But that's just a wild guess. Anyway ... some details:

-- brand new card, purchased a week ago. Box says model G55-MDDE32F pci-express x1 32MB DDR dual output standard height half-width
-- lspci identifies it as Matrox Graphics, Inc. MGA G550 AGP (rev 01) Note that this means its slightly mis-identified: its NOT AGP, its a PCI-E 1x card!
-- It does output vga resolution during boot; displays standard text 24 rows x 80 columns just fine (and displays BIOS menus just fine, etc).
-- monitor and cable work fine when attached to different graphics card, even when driven at high-res (1920x1080).
-- tested both connectors; not output on either.
-- The xorg.0.log file shows no errors; it does detect monitor just fine, and autoconfigure just fine.
-- Forcing xorg.conf to use a lower resolution does not solve problem
-- X-server appears to be running fine, based on sound feedback. I was able to (blindly) log in, start web browser, go to youtube, and start a video playing. The sound-track came out the audio; screen was black. In other words, X11 itself seems to be running just fine, but the output is completely blanked.
-- Above reports were for Ubuntu karmic; tried upgrading to Oneiric, no improvement.

Will attach xorg.conf and xorg.0.log shortly

[linas (linasvepstas)]

xorg.conf file

[linas (linasvepstas)]

Log file from a failed session. Several notes about this file:

-- you will see some vague evidence for another graphics adapter in the log; its an nvidia at pci 4:0:0 Please note the mga problem occurs even when the nvidia card is removed from the system ... I checked!

-- Note that the xorg.conf file uses a stanza indicating that VideoRam 32768 This is required; without this, I get the following oddball behaviour: early device probe prints out that the card has only 2 MB video ram; then just a little while later, the video bios is read, and reports that the card has 32 MB ram. Later still, during video mode selection, it rejects all modes requiring anything more than 2MB (i.e. it rejects almost all modes) i.e. it behaves as if it "forgot" that the 32 MB was reported.

[linas (linasvepstas)]

BTW, If I use a DVI-to-VGA adapter, and run a vga cable to the monitor, then one output port shows fine pink& black vertical bars; the other port shows all black (and the monitor thinks the all-black screen does have valid video out, at 1600x1200) (the pink-bars vga out is also 1600x1200). By contrast, running a DVI cable simply results in the monitor complaining: "no signal". Yes, I have tried restarting Xserver before & after the cable switches.

[linas (linasvepstas)]

BTW, this is on an amd64 system (i.e. 64-bit binaries) perhaps there's some 32->64-bit bug?

[Tormod Volden (tormodvolden)]

Please run "apport-collect 900111" to attach more information. It would be nice to also have the "xrandr --verbose" output from the "blind" session.

BTW, I saw this in the log. Do you have the proprietary nvidia OpenGL libraries installed?
[ 81272.142] (EE) Failed to initialize GLX extension (Compatible NVIDIA X driver not found)

BTW2: The misleading lspci information would need to be rectified at http://pciids.sourceforge.net/
I might look into this some day, since I also have some other IDs to fix up. Is there any model identifier on the card itself, or in the BIOS messages?

[Tormod Volden (tormodvolden)]

Please file a separate bug for the memory size detection. The driver does all kinds of trickery to figure out the VRAM size, and this should be possible to get to work also in your case.

[linas (linasvepstas)]

apport-collect fails: I get all the way through a web session, where I loig into launchpad openid via lynx, authorize apport to collect the data, and then get this:

No REFERER Header

   Launchpad requires a REFERER header to perform this action. There is no
   REFERER header present. This can be caused by configuring your browser
   to block REFERER headers.

No clue how to fix that by hand, so I try apport-bugs instead. That also fails, here's the full session:

root@blackspot:~# apport-bug

*** What kind of problem do you want to report?

Choices:
  1: Display (X.org)
  2: External or internal storage devices (e. g. USB sticks)
  3: Security related problems
  4: Sound/audio related problems
  5: Other problem
  C: Cancel
Please choose (1/2/3/4/5/C): 1

*** Collecting problem information

The collected information can be sent to the developers to improve the
application. This might take a few minutes.

*** What display problem do you observe?

Choices:
  1: I don't know
  2: Freezes or hangs during boot or usage
  3: Crashes or restarts back to login screen
  4: Resolution is incorrect
  5: Shows screen corruption
  6: Performance is worse than expected
  7: Fonts are the wrong size
  8: Other display-related problem
  C: Cancel
Please choose (1/2/3/4/5/6/7/8/C): 8
.No packages found matching xorg.
No packages found matching xorg.
...........

*** Problem in xorg

The problem cannot be reported:

The running kernel is not an Ubuntu kernel

Press any key to continue...

root@blackspot:~#

FWIW, I'm not running an ubuntu kernel, for two reasons:
1) I'm a kernel developer, this is my area of expertise, and
2) Ubuntu has never-ever done RAID correctly, and so I've had to run a custom kernel just to get RAID to work.

----------
Yes, I have proprietary nvidia drivers installed. I explained this above. Physically removing the nvidia card does not alter the behaviour. I am planning on eventually having a multi-head setup, running both the matrox and the nvidia at the same time; first step is to get mga to work. (nvidia works fine)

Will post xrandr output shortly; I've looked at this before, it looked fine.

[linas (linasvepstas)]

attachment has xrandr output. First few lines are these:

linas@blackspot: ~ $ xrandr --verbose
xrandr: Failed to get size of gamma for output default
Screen 0: minimum 400 x 300, current 1920 x 1080, maximum 1920 x 1080
default connected 1920x1080+0+0 (0x4c) normal (normal) 0mm x 0mm
 Identifier: 0x4b
 Timestamp: 5344371
 Subpixel: horizontal rgb
 Clones:
 CRTC: 0
 CRTCs: 0
 Transform: 1.000000 0.000000 0.000000
             0.000000 1.000000 0.000000
             0.000000 0.000000 1.000000
            filter:
  1920x1080 (0x4c) 124.4MHz *current
        h: width 1920 start 0 end 0 total 1920 skew 0 clock 64.8KHz
        v: height 1080 start 0 end 0 total 1080 clock 60.0Hz
  1680x1050 (0x4d) 105.8MHz
        h: width 1680 start 0 end 0 total 1680 skew 0 clock 63.0KHz
        v: height 1050 start 0 end 0 total 1050 clock 60.0Hz

[linas (linasvepstas)]

p.s. as a kernel/toolchain/glibc developer who once used to do X11 devel long long ago, I'm willing to build from source, in order to test.

[Tormod Volden (tormodvolden)]

Here is your chance to get back into X development! :) Seriously, having the hardware at hand is a huge advantage. Meanwhile I have found a similar bug report here: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=546836 where it appears to be a regression since the old XFree86 server. I have a feeling some of these regressions boil down to mode calculation being done differently. Maybe the card is driving your monitor out of range? Did you already try other screen resolutions and refresh frequencies? You can use "xrandr --output default --mode 0xXX" where you can find XX in the xrandr --verbose listing. Does your screen come back when switching to a virtual console?

[Tormod Volden (tormodvolden)]

Oh, just saw "Forcing xorg.conf to use a lower resolution does not solve problem" now. Still, please cycle through all the xrandr modes.

[linas (linasvepstas)]

The bug I report her appears to be old and well-known to xorg/freedesktop See
https://bugs.freedesktop.org/show_bug.cgi?id=24966 Will attempt to download and debug source code...