kvm-devel

[kvm-devel] KVM console dying

[Marcelo T. <mto...@re...>]

From: Alan Pevec <ap...@re...>

> - add serial console, workaround for F9 livecd KVM guest dying with 
> standard console only.
>  VNC console will go blank but node will continue to boot

With only console=tty qemu-kvm dies when, AFAICT from udevdebug output, start_udev is processing console rules.

Here is backtrace:

Program received signal SIGSEGV, Segmentation fault.

[Switching to Thread -1208993168 (LWP 16163)]

0x0095337d in memmove () from /lib/libc.so.6

Missing separate debuginfos, use: debuginfo-install SDL.i386 alsa-lib.i386 glibc.i686 gnutls.i386 libgcrypt.i386 libgpg-error.i386 zlib.i386

(gdb) bt

#0  0x0095337d in memmove () from /lib/libc.so.6

#1  0x080b38ae in vnc_copy (ds=0x8181060, src_x=64, src_y=416, dst_x=64, 

    dst_y=400, w=72, h=16) at vnc.c:465

#2  0x080947ec in cirrus_bitblt_start (s=0xa2181e0)

    at /usr/src/debug/kvm-60/qemu/hw/cirrus_vga.c:760

#3  0x08095008 in cirrus_hook_write_gr (s=0xa2181e0, reg_index=2534224968, 

    reg_value=2) at /usr/src/debug/kvm-60/qemu/hw/cirrus_vga.c:1064

#4  0x08095bb9 in vga_ioport_write (opaque=0xa2181e0, addr=975, val=2)

    at /usr/src/debug/kvm-60/qemu/hw/cirrus_vga.c:2958

#5  0x080d3277 in kvm_outw (opaque=0x0, addr=3072, data=<value optimized out>)

    at /usr/src/debug/kvm-60/qemu/qemu-kvm.c:478

#6  0x0814b643 in kvm_run (kvm=0xa0de028, vcpu=0) at libkvm.c:719

#7  0x080d2bf2 in kvm_cpu_exec (env=0xa1085e0)

    at /usr/src/debug/kvm-60/qemu/qemu-kvm.c:143

#8  0x080d36d2 in kvm_main_loop_cpu (env=0xa1085e0)

    at /usr/src/debug/kvm-60/qemu/qemu-kvm.c:343

#9  0x08057cf8 in main (argc=20, argv=0xbfc42914)

    at /usr/src/debug/kvm-60/qemu/vl.c:7817

#10 0x008f8390 in __libc_start_main () from /lib/libc.so.6

#11 0x0804ce21 in _start ()


VNC server bug? Can QEMU/KVM experts have a look? Any other hints?

My host is F8, kernel-2.6.24.4-64.fc8 kvm-60-3.fc8 libvirt-0.4.1-3.fc8.

Thanks!

Re: [kvm-devel] KVM console dying

[Anthony L. <an...@co...>]

Marcelo Tosatti wrote:
> From: Alan Pevec <ap...@re...>
>
>   
>> - add serial console, workaround for F9 livecd KVM guest dying with 
>> standard console only.
>>  VNC console will go blank but node will continue to boot
>>     
>
> With only console=tty qemu-kvm dies when, AFAICT from udevdebug output, start_udev is processing console rules.
>
> Here is backtrace:
>
> Program received signal SIGSEGV, Segmentation fault.
>
> [Switching to Thread -1208993168 (LWP 16163)]
>
> 0x0095337d in memmove () from /lib/libc.so.6
>
> Missing separate debuginfos, use: debuginfo-install SDL.i386 alsa-lib.i386 glibc.i686 gnutls.i386 libgcrypt.i386 libgpg-error.i386 zlib.i386
>   

The info that's needed is what the size of the vnc frame buffer is.  My 
suspicion is that we're racing.  A VCPU IO operation will trigger the 
bit blit which will cause this vnc_copy path to hit.  However, the VNC 
server is not thread safe so if the IO thread is running simultaneously, 
very bad things could happen.

Is this with standard KVM or your lock break-up patches?

Regards,

Anthony Liguori

Re: [kvm-devel] KVM console dying

[Marcelo T. <mto...@re...>]

On Mon, Apr 21, 2008 at 01:24:00PM -0500, Anthony Liguori wrote:
> Marcelo Tosatti wrote:
> >From: Alan Pevec <ap...@re...>
> >
> >  
> >>- add serial console, workaround for F9 livecd KVM guest dying with 
> >>standard console only.
> >> VNC console will go blank but node will continue to boot
> >>    
> >
> >With only console=tty qemu-kvm dies when, AFAICT from udevdebug output, 
> >start_udev is processing console rules.
> >
> >Here is backtrace:
> >
> >Program received signal SIGSEGV, Segmentation fault.
> >
> >[Switching to Thread -1208993168 (LWP 16163)]
> >
> >0x0095337d in memmove () from /lib/libc.so.6
> >
> >Missing separate debuginfos, use: debuginfo-install SDL.i386 alsa-lib.i386 
> >glibc.i686 gnutls.i386 libgcrypt.i386 libgpg-error.i386 zlib.i386
> >  
> 
> The info that's needed is what the size of the vnc frame buffer is.  My 
> suspicion is that we're racing.  A VCPU IO operation will trigger the 
> bit blit which will cause this vnc_copy path to hit.  However, the VNC 
> server is not thread safe so if the IO thread is running simultaneously, 
> very bad things could happen.
> 
> Is this with standard KVM or your lock break-up patches?

Standard KVM.