Bug 84689

Please do a binary search of last builds to narrow down the single release
that introduces the problem for ya.

17.1 works, 18.0 doesn't

The machine with the crashing server has a Matrox G400 video card, by the way.

Too weird.

So if I set the inittab line to run:
sh -c '/etc/X11/X -query othermachine'
then it will segfault.

But this works fine:
sh -c 'cd /;/etc/X11/X -query othermachine'

About the best stack trace I'm going to be able to get is this I think:

#0  0x4011e985 in ?? ()
#1  0x0876c646 in ?? ()
#2  0x08782ea7 in ?? ()
#3  0x08782bc3 in ?? ()
#4  0x08085dd6 in DGAShutdown () at xf86DGA.c:496
#5  0x0806e87e in ddxGiveUp ()
#6  0x08100512 in AbortServer () at utils.c:436
#7  0x08102060 in FatalError (
    f=0x81ed6c0 "Caught signal %d.  Server aborting\n") at utils.c:1421
#8  0x0808861e in xf86SigHandler (signo=11) at xf86Events.c:1186
#9  <signal handler called>
#10 0x4011e985 in ?? ()
#11 0x0876c4a4 in ?? ()
#12 0x08782771 in ?? ()
#13 0x080becec in CMapCloseScreen (i=0, pScreen=0x87a4aa8) at xf86cmap.c:215
#14 0x0809734f in DPMSClose (i=0, pScreen=0x87a4aa8) at xf86DPMS.c:100
#15 0x0830a800 in ?? ()
#16 0x4011a4df in ?? ()
#17 0x08193b94 in RRCloseScreen (i=0, pScreen=0x87a4aa8) at randr.c:159
#18 0x08096f8f in xf86RandRCloseScreen (index=0, pScreen=0x87a4aa8)
    at xf86RandR.c:235
#19 0x080b79f3 in VidModeClose (i=0, pScreen=0x87a4aa8) at xf86VidMode.c:100
#20 0x0818f9e8 in LbxCloseScreen (i=0, pScreen=0x87a4aa8) at lbxcmap.c:159
#21 0x081a2775 in AnimCurCloseScreen (index=0, pScreen=0x87a4aa8)
    at animcur.c:126
#22 0x080e16e9 in main (argc=3, argv=0xbfffea74, envp=0xbfffea84) at main.c:459
#23 0x420154a4 in __libc_start_main () from /lib/tls/libc.so.6

The only other thing I've managed to capture is this fragment of strace output.
 Running X under strace to start with ends up looping with signals, so I had to
attach it at the instant before it crashed; not easy.

write(0, "[drm] unmapping 8192 bytes of SA"..., 61) = 61
munmap(0x4001b000, 8192)                = 0
fstat64(7, {st_mode=S_IFCHR|0666, st_rdev=makedev(226, 0), ...}) = 0
fstat64(7, {st_mode=S_IFCHR|0666, st_rdev=makedev(226, 0), ...}) = 0
close(7)                                = 0
--- SIGSEGV (Segmentation fault) @ 0 (0) ---

I put this in section 'device' (mga driver):

Option "HWCursor" "off"

That makes the problem go away.  Some hwcursor bug I guess.

Created attachment 90268 [details]
XFree86-revert-some.patch

This patch (which reverts a change in 0217->0218) seems to make the problem go
away, unless I tested wrong.

This patch, which was checked into CVS by Alan, should fix this problem.

   939. Check pScrn->vtSema before calling xf86SetCursor() from
        xf86CursorCloseScreen().  This avoids a segfault at exit with some
        drivers (Alan Hourihane).

--- xc/programs/Xserver/hw/xfree86/ramdac/xf86Cursor.c:1.18     Tue Feb 18
10:42:13 2003
+++ xc/programs/Xserver/hw/xfree86/ramdac/xf86Cursor.c  Fri Feb 21 17:40:33 2003
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/ramdac/xf86Cursor.c,v 1.17tsi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/ramdac/xf86Cursor.c,v 1.18
2003/02/18 15:42:13 tsi Exp $ */
 
 #include "xf86.h"
 #include "xf86_ansic.h"
@@ -122,7 +122,7 @@
     xf86CursorScreenPtr ScreenPriv =
        pScreen->devPrivates[xf86CursorScreenIndex].ptr;
 
-    if (ScreenPriv->isUp)
+    if (ScreenPriv->isUp && pScrn->vtSema)
        xf86SetCursor(pScreen, NullCursor, ScreenPriv->x, ScreenPriv->y);
 
     pScreen->CloseScreen = ScreenPriv->CloseScreen;

Setting to MODIFIED, please test XFree86-4.2.99.902-20030223.0
build and set to RAWHIDE if fixed or ASSIGNED if broken still.

Yes, seems to fix it here.  Thanks.