Bug 136677 - radeon: X dies of SIGFPE
Summary: radeon: X dies of SIGFPE
Status: RESOLVED WONTFIX
Alias: None
Product: SUSE Linux 10.1
Classification: openSUSE
Component: X.Org (show other bugs)
Version: Alpha 3plus
Hardware: Other Other
: P5 - None : Normal (vote)
Target Milestone: ---
Assignee: Egbert Eich
QA Contact: E-mail List
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2005-12-02 13:49 UTC by Andreas Schwab
Modified: 2008-04-19 06:06 UTC (History)
2 users (show)

See Also:
Found By: Other
Services Priority:
Business Priority:
Blocker: ---
Marketing QA Status: ---
IT Deployment: ---


Attachments
X config generated by SaX2 (7.12 KB, text/plain)
2005-12-02 13:50 UTC, Andreas Schwab
Details
hand-crafted X config (9.68 KB, text/plain)
2005-12-02 14:20 UTC, Andreas Schwab
Details
Xorg.0.log (73.92 KB, text/plain)
2007-05-08 17:11 UTC, Andreas Schwab
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Andreas Schwab 2005-12-02 13:49:04 UTC
In radeon_driver.c:
5530            /* FIXME: When we move to dynamic allocation of back and depth
5531             * buffers, we will want to revisit the following check for 3
5532             * times the virtual size of the screen below.
5533             */
5534            int  width_bytes = (pScrn->displayWidth *
5535                                info->CurrentLayout.pixel_bytes);
5536            int  maxy        = info->FbMapSize / width_bytes;

pScrn->displayWidth is zero, thus you'll get a div by 0.  This appears to be dependent on the config, this happens with the one generated by SaX2, but not with a hand-crafted one derived from X -configure.
Comment 1 Andreas Schwab 2005-12-02 13:50:07 UTC
Created attachment 59701 [details]
X config generated by SaX2
Comment 2 Stefan Dirsch 2005-12-02 13:59:21 UTC
Hmm ... could you attach the hand-crafted as well? SaX2 seems to create a dualhead MergedFB configuration for 4 monitors, which looks correct to me.
Comment 3 Andreas Schwab 2005-12-02 14:20:12 UTC
Created attachment 59704 [details]
hand-crafted X config
Comment 4 Stefan Dirsch 2005-12-02 14:33:00 UTC
Your hand-crafted one seems to be a non mergedfb configuration. So you connected your 2 monitors to different boards, right?
Comment 5 Andreas Schwab 2005-12-02 15:37:58 UTC
Boards = cards?  Yes.
Comment 6 Stefan Dirsch 2005-12-02 15:45:54 UTC
Yes, boards=cards. Ok. This makes sense to me. So I think we see a MergedFB problem here. :-( I would like to know, whether it only occurs with two boards, since I didn't see it yet with one board active (at least not on i386/x86_64).
Comment 7 Stefan Dirsch 2005-12-02 15:47:23 UTC
Removing 

  Screen       "Screen[1]" RightOf "Screen[0]"

in the SaX2 generated config and connecting the two monitors to the first card should do the trick.
Comment 8 Andreas Schwab 2005-12-08 12:27:23 UTC
Is there a way to tell sax not to enable the MergedFB option?
Comment 9 Stefan Dirsch 2005-12-08 13:16:15 UTC
Could you answer my question first?
Comment 10 Andreas Schwab 2005-12-08 13:26:40 UTC
Do you mean comment #7?
It does not help.  While the server does not crash any more, it doesn't display anything either.
Comment 11 Stefan Dirsch 2005-12-08 14:01:17 UTC
Yes, comment #7. Thanks. I think you could use the Radeon_DualHead profile instead.

sax2 -r -b /usr/share/sax/profile/Radeon_DualHead or something like this.
Comment 12 Andreas Schwab 2005-12-08 16:12:58 UTC
Doesn't help either.

(WW) RADEON(0): Failed to detect secondary monitor, MergedFB/Clone mode disabled
(II) RADEON(0): MergedFB mode forced on.
(WW) RADEON(2): Failed to detect secondary monitor, MergedFB/Clone mode disabled
(II) RADEON(2): MergedFB mode forced on.
Comment 13 Stefan Dirsch 2005-12-08 16:30:31 UTC
I hoped that SaX2 would create a config similar to your handcrafted one, but this seems to be wrong. Anyway, Marcus knows more about this multihead config stuff than I do.
Comment 14 Stefan Dirsch 2006-09-02 19:06:43 UTC
I suggest to check this again with openSUSE >= 10.2 once the X Server works on openSUSE again.
Comment 15 Stefan Dirsch 2006-10-23 12:53:25 UTC
(In reply to comment #14)
> I suggest to check this again with openSUSE >= 10.2 once the X Server works
> on openSUSE again.
Not sure if we're already prepared for this though. But it might be worth a try.

Comment 16 Stefan Dirsch 2006-12-21 17:58:24 UTC
Andreas, did you check this with openSUSE 10.2 again?
Comment 17 Andreas Schwab 2007-01-18 13:53:24 UTC
SaX does not work at all.
Comment 18 Stefan Dirsch 2007-04-02 19:59:08 UTC
Andreas, any improvements with STABLE or is X.Org still broken on STABLE-IA64?
Comment 19 Egbert Eich 2007-05-07 09:41:48 UTC
Stefan, is this the randr 1.2 enabled driver?
Comment 20 Stefan Dirsch 2007-05-07 10:30:57 UTC
No. BTW, RANDR 1.2 for radeon driver still lives in a separate git branch.
Comment 21 Andreas Schwab 2007-05-07 11:12:00 UTC
SaX2 is not able to detect the monitor and only uses 800x600.  Also I don't see a way to make it use both graphics cards.
Comment 22 Egbert Eich 2007-05-07 13:37:56 UTC
There are still instances where something gets devided by the screen width. I don't see any sanity checks in the code so if this is gone it's gone because some upper layer code takes care of this.
Andreas, would you please test this with the config attached in comment #1?
I've got a similar machine I think but I'd need to update it first and I'm not set up for this right now.
Comment 23 Andreas Schwab 2007-05-08 10:55:38 UTC
The config no longer works at all.
Comment 24 Stefan Dirsch 2007-05-08 11:23:41 UTC
Hmm. Could you attach the logfile for this config? Thanks.
Comment 25 Andreas Schwab 2007-05-08 17:11:40 UTC
Created attachment 138348 [details]
Xorg.0.log
Comment 26 Stefan Dirsch 2007-05-12 10:42:25 UTC
JFYI, Matthias. This is a bugreport, which is assigned to Egbert/me or with Egbert/me in CC or reported by Egbert/me.
Comment 27 Egbert Eich 2008-03-31 18:40:42 UTC
The radeon driver has been reworked heavily. 
Andreas, is this bug still an issue?
Stefan do you know  if this server/driver in question of any relevance for an enterprise product?
Comment 28 Stefan Dirsch 2008-03-31 20:09:45 UTC
Well, SLES10 is based on SUSE LINUX 10.1, so it might be relevant. But it seems the issue only occurs if you configure two seperate ATI gfx cards each one using MergedFB mode ==> 4 heads.

For SLES11 we'll switch to the new radeon driver with RANDR >= 1.2 support, which obsoletes MergedFB, so *this* issue will disappear.

I suggest to close this bugreport as WONTFIX. I'm not aware of anyone running
into this problem than Andreas, and since he's using a valid workaround ...

I'm rather sure we can still reproduce this issue on our SLI test machine with
two ATI cards - at least on SLES10. But is it really that important?

> Andreas, is this bug still an issue?
I'm not sure, if Andreas can answer this question easily at the moment. STABLE and X might not be usable at all on IA64 at this time. :-(
Comment 29 Egbert Eich 2008-04-12 12:04:29 UTC
I would like to know if this should eventually be fixed for the next SP for SLES10.
If not I will close it with WONTFIX.
Comment 30 Stefan Dirsch 2008-04-19 06:06:47 UTC
Looks like there's no need to fix it still for SLES10. ==> WONTFIX