Bug 78002

Hmm, I broke something in the saving of malloc errors then. Is there anything in the crashing Xorg.0.log or Xorg.0.log.old?

(In reply to comment #1)
> Is there
> anything in the crashing Xorg.0.log or Xorg.0.log.old?

The last entry of the .old (which had a change date matching the previous X session) has those last entrys - which are from the startup:

> [    23.934] (II) config/udev: Adding input device PC Speaker (/dev/input/event11)
> [    23.934] (II) No input driver specified, ignoring this device.
> [    23.934] (II) This device may have been added with another device file.

So, no, there is nothing relating to the X restart.

> $ less Xorg.0.log.old | grep -i cursor
> [    22.221] (--) intel(0): Using a maximum size of 64x64 for hardware cursors
> [    22.316] (II) intel(0): HW Cursor enabled

Nothing special.

Is the stderr captured anywhere? Usually as xdm.log or gdm/*.log etc.

I am using KDM. `journalctl _SYSTEMD_UNIT=kdm.service`    did not output any log entries. And in /var/log/kdm.log there is not anything interesting.  Not sure where the stderr of kdm and its childs go to... I will try without kdm:


With a pure startx 2> stderr.log (no kdm not enabled in systemd) and a exec xterm X crashes on the 
third re-entry of the mouse cursor. Content generated by the crash:

> xterm: xinit: connection to X server lost
> fatal IO error 11 (Die Ressource ist zur Zeit nicht verfügbar) or KillClient on X server ":0"

With lxde instead of xterm it even crashes at the first reappear of the mouse cursor:

> XIO:  fatal IO error 11 (Resource temporarily unavailable) on X server ":0"
>       after 31 requests (31 known processed) with 0 events remaining.
> xinit: connection to X server lost
> 
> waiting for X server to shut down XIO:  fatal IO error 11 (Die Ressource ist zur Zeit nicht 
> verfügbar) on X server ":0"
>       after 3514 requests (3511 known processed) with 0 events remaining.
> pcmanfm: Fatal IO error 11 (Die Ressource ist zur Zeit nicht verfügbar) on X server :0.

By the way, X does not "restart" (as I first wrote in the bug title) when not using KDM. Apparently the restarting is a feature of KDM. Being just on tty1 with startx it crashes and makes the computer not respond to any input. Accessing by ssh (and rebooting) works.

However, I guess those stderrs are not really useful for you. Please tell me how to investigate better.

Hmm, if we can't see a reason for the crash in either the log file or stderr, we need to attach gdb. This is easiest with a second machine and sshing in. So far, I haven't spotted anything in either code review or running with valgrind - but I still haven't tried hooking up a VGA monitor to the 855gm.

(In reply to comment #5)
> Hmm, if we can't see a reason for the crash in either the log file or
> stderr, we need to attach gdb. This is easiest with a second machine and
> sshing in. 

Sure, this is really no problem, I also tried it before, but as soon as I attach gdb (via PID specification) X's execution seems to stop (and I need it to continue to run to trigger the error...). Probably I just need to tell to "continue" or something but I did not find anything in the man page (quickly reading). If you could tell me how to attach and continue to run X and then backtracing ...  Hmm, I just found http://visualgdb.com/gdbreference/commands/continue which tells me that I just need to issue the command "continue". Does it work this way (we have no breakpoints).

> but I still haven't tried hooking up a VGA monitor to the 855gm.

Well, I guess it is related to the different monitor sizes, so you likely only can reproduce it with a bigger / smaller external VGA monitor.

Right, you need to hit continue. So connect gdb, using gdb --pid=`pidof Xorg` (or perhaps pidof X), then enter 'c' when it finishes loading the symbols and then trigger the error. When X dies, gdb will hopefully capture the error and present a command prompt again. Type 'bt'.

To get a good backtrace, typically requires the debugging symbols to be installed.

Another thing to check is whether there is a tell-tale in dmesg for why X died.

nothing interesting/new in dmesg. I have no debugging symbols installed but I think it is enough:

Tested with your newest version 2a993c8. 

> Program received signal SIGSEGV, Segmentation fault.
> __sna_create_cursor (sna=0xb6aec000) at sna_display.c:3112
> 3112		c->alloc = ALIGN(size, 4096);

Also see attachment.

Created attachment 98129 [details]
X crash message in gdb and backtrace

Ah, that makes sense at least. It should be impossible... Do you mind applying the

index 1520533..854ee55 100644
--- a/src/sna/sna_display.c
+++ b/src/sna/sna_display.c
@@ -87,7 +87,7 @@ union compat_mode_get_connector{
 #define DEFAULT_DPI 96
 #endif
 
-#if 0
+#if 1
 #define __DBG(x) ErrorF x
 #else
 #define __DBG(x)

debugging patch and attaching the tail of the Xorg.0.log? I need to work out why we end up using more cursors than I preallocate.

hmm, the patch does not work (if applied to version 2a993c8aa9e8594c32d5e67329b0dbed0d92c761 or 0b23011c27736d0ae2b33d8ea147c16b909baa57)

> $ patch sna_display.c /tmp/debugpatch
> patching file sna_display.c
> patch unexpectedly ends in middle of line
> Hunk #1 FAILED at 87.
> 1 out of 1 hunk FAILED -- saving rejects to file sna_display.c.rej

But I could apply the one from https://bugs.freedesktop.org/show_bug.cgi?id=77351#c4

> patch unexpectedly ends in middle of line
> Hunk #1 succeeded at 87 with fuzz 1.

Trying this one...

Created attachment 98137 [details]
gdb for comment 12 ...

[   107.955] (II) intel(0): SNA compiled from 2.99.911-113-g0b23011

Created attachment 98138 [details]
part of Xorg log while the debug patch was active - for comment 12

likely uninteresting info: ... actually this was Xorg.1.log because I did a startx on tty1 while there was kdm running on tty7.

Ugh. It's the cursor sharing avoidance for the pwrite paths.

commit 94e39323772ef6561efcc0620f67cabd2462a0d0
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Apr 29 09:02:50 2014 +0100

    sna: Recycle physical cursors
    
    A side-effect of the workaround for incoherent physical cursors is that
    we never reused a cursor after disabling. As such moving the cursor off
    the pipe and back on would eventually consume all the preallocated
    structs leading to a segfault.
    
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78002
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Thanks! Confirming, seems to be fixed in 

> intel(0): SNA compiled from 2.99.911-114-g94e3932