Bug 1064466

Created attachment 862417 [details]
File: backtrace

Created attachment 862418 [details]
File: cgroup

Created attachment 862419 [details]
File: core_backtrace

Created attachment 862420 [details]
File: dso_list

Created attachment 862421 [details]
File: environ

Created attachment 862422 [details]
File: limits

Created attachment 862423 [details]
File: maps

Created attachment 862424 [details]
File: proc_pid_status

Created attachment 862425 [details]
File: var_log_messages

Created attachment 862426 [details]
File: sosreport.tar.xz

Actually, this happens only when the guest in virt-manager is configured to use the cirrus video device.  Also, in that case, the X server in the guest seems to fallback to vesa since there's no cirrus package included in RHEL7 any more (not sure if it's by intention or not).

Here's a more precise reproducer:

1. Install a guest RHEL7 in virt-manager and use cirrus video and VNC
2. Boot it up and log in a GNOME session
3. Set the display resolution in control-center to the highest available (1280x1024 in my case)
4. Reboot and log in again
5. Change the resolution to the lowest available (800x600 in my case)

I can't reproduce with the qxl + spice combination.

Another user experienced a similar problem:

Following reproducer in other bug[1], I have

*   installed tigervnc packages on two machines (A, B),
*   changed keyboard layout and added some X11 config files
    on machine A
*   turned off firewall
*   and attempted to connect from machine B using
        $ vncviewer <ip-address>
    called from GNOME session

(Please visit the link for full reproducer.)

  [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1054118#c5

The connection was closed by B after showing the screen. Then I found this
crash report on machine B.

Note that the above reproducer is written for RHEL6.5 and different version of
tigervnc.

reporter:       libreport-2.1.11
backtrace_rating: 4
cmdline:        /usr/bin/Xorg :0 -background none -verbose -auth /run/gdm/auth-for-gdm-ILBJkT/database -seat seat0 -nolisten tcp vt1
crash_function: OsAbort
executable:     /usr/bin/Xorg
kernel:         3.10.0-86.el7.x86_64
open_fds:       
package:        xorg-x11-server-Xorg-1.15.0-3.el7
reason:         Xorg killed by SIGABRT
runlevel:       N 5
type:           CCpp
uid:            0

(In reply to Michal Domonkos from comment #12)
> Actually, this happens only when the guest in virt-manager is configured to
> use the cirrus video device.  Also, in that case, the X server in the guest
> seems to fallback to vesa

You mean modesetting.

> since there's no cirrus package included in RHEL7
> any more (not sure if it's by intention or not).

Intentional.

This is proving to be infuriatingly difficult to reproduce under gdb.  However, if I just go by the backtrace:

#10 ms_shadowUpdate32to24 (pScreen=<optimized out>, pBuf=<optimized out>) at sh3224.c:134
        damage = <optimized out>
        pShadow = <optimized out>
        nbox = 0
        pbox = 0x19a0188
        shaStride = 800
        shaBpp = 32
        x = <optimized out>
        y = <optimized out>
        w = 40
        h = 39
        winSize = 2400
        shaBase = 0x14b5c50
        shaLine = 0x16ada78
        winBase = 0x0
        winLine = 0x7f38920eee9e ""

winBase of 0 is definitely not supposed to happen, so I'm not really sure how we'd get here.

I'm not sure the backtrace is accurate though, since nbox = 0 is also invalid,

also the src in the next frame looks kinda valid, though who knows since its 64-bit.

I've even tried reproducing this with the vnc bits installed in case it makes it easier and its not happening for me at all.

Low impact, moving to 7.1

Reassign to xorg-x11-server, which has replaced -modesetting in 7.2.

Closing this, we haven't been able to reproduce it for years. Please reopen if this is still an issue.