[Qemu-devel] How to reliably start a bare QEMU target to query capabilities via QMP

[Qemu-devel] How to reliably start a bare QEMU target to query capabilities via QMP

[Daniel P. Berrange]

I've been adapting libvirt to use to the various new QMP commands to
query QEMU's capabilities, instead of the hated -help parsing. Obviously
the critical part of this is being able to reliably start a bare QEMU
process with no actual guest OS configured (no disks, no kernel, etc)
and talk to its monitor. I hadn't anticipated problems since I only
tested with x86_64 / i386 most of the time, which work well in this
respect. Currently I am doing

  # $QEMU_BINARY -S \
         -no-user-config \
         -nodefconfig \
         -nodefaults \
         -nographic \
         -qmp stdio


This works for about 50% of the QEMU targets:

    qemu-system-alpha
    qemu-system-i386
    qemu-system-lm32
    qemu-system-ppc
    qemu-system-ppc64
    qemu-system-s390x
    qemu-system-sparc
    qemu-system-sparc64
    qemu-system-x86_64
    qemu-system-xtensa
    qemu-system-xtensaeb

 but fails for the other 50% of targets:

    qemu-system-arm
    qemu-system-cris
    qemu-system-m68k
    qemu-system-mips
    qemu-system-mips64
    qemu-system-mips64el
    qemu-system-mipsel
    qemu-system-or32
    qemu-system-ppcemb
    qemu-system-sh4
    qemu-system-sh4eb
    qemu-system-unicore32


With the failing targets i see the following kinds of errors:


 $ qemu-system-arm -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Kernel image must be specified

 $ qemu-system-cris -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Kernel image must be specified

 $ qemu-system-m68k -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Kernel image must be specified

 $ qemu-system-mips -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait qemu: Could not load MIPS bios 'mips_bios.bin', and no -kernel argument was specified

 $ qemu-system-mipsel -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 qemu: Could not load MIPS bios 'mipsel_bios.bin', and no -kernel argument was specified

 $ qemu-system-or32 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Can't create serial device, empty char device

 $ qemu-system-ppcemb -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Unable to find PowerPC CPU definition

 $ qemu-system-sh4 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Initializing CPU
 Allocating ROM
 Allocating SDRAM 1
 Allocating SDRAM 2
 shix_init: load BIOS 'shix_bios.bin'
 ret=-1
 qemu: could not load SHIX bios 'shix_bios.bin'

 $ qemu-system-sh4eb -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 Initializing CPU
 Allocating ROM
 Allocating SDRAM 1
 Allocating SDRAM 2
 shix_init: load BIOS 'shix_bios.bin'
 ret=-1
 qemu: could not load SHIX bios 'shix_bios.bin'

 $ qemu-system-unicore32 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
 qemu-system-unicore32: /home/berrange/src/virt/qemu/hw/unicore32/../puv3.c:81: puv3_load_kernel: Assertion `kernel_filename != ((void *)0)' failed.
 Aborted


Can we "fix" them in some way to not require the kernel ?

Do we have to go down the route of adding some sort of "-no-vm" flag to
explicitly say we don't care about any of the VM setup parts ? This would
be more like what my old patches did which mapped monitors commands
onto '-query-XXXXX' command line args, bypassing VM setup, but still
using normal QMP monitor interaction.

Any other suggestions on how to reliably get a QMP monitor to a target,
without any VM config. ?

Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

Re: [Qemu-devel] How to reliably start a bare QEMU target to query capabilities via QMP

[Anthony Liguori]

"Daniel P. Berrange" <berrange@redhat.com> writes:

> I've been adapting libvirt to use to the various new QMP commands to
> query QEMU's capabilities, instead of the hated -help parsing. Obviously
> the critical part of this is being able to reliably start a bare QEMU
> process with no actual guest OS configured (no disks, no kernel, etc)
> and talk to its monitor. I hadn't anticipated problems since I only
> tested with x86_64 / i386 most of the time, which work well in this
> respect. Currently I am doing
>
>   # $QEMU_BINARY -S \
>          -no-user-config \
>          -nodefconfig \
>          -nodefaults \
>          -nographic \
>          -qmp stdio
>
>
> This works for about 50% of the QEMU targets:
>
>     qemu-system-alpha
>     qemu-system-i386
>     qemu-system-lm32
>     qemu-system-ppc
>     qemu-system-ppc64
>     qemu-system-s390x
>     qemu-system-sparc
>     qemu-system-sparc64
>     qemu-system-x86_64
>     qemu-system-xtensa
>     qemu-system-xtensaeb
>
>  but fails for the other 50% of targets:
>
>     qemu-system-arm
>     qemu-system-cris
>     qemu-system-m68k
>     qemu-system-mips
>     qemu-system-mips64
>     qemu-system-mips64el
>     qemu-system-mipsel
>     qemu-system-or32
>     qemu-system-ppcemb
>     qemu-system-sh4
>     qemu-system-sh4eb
>     qemu-system-unicore32
>
>
> With the failing targets i see the following kinds of errors:
>
>
>  $ qemu-system-arm -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Kernel image must be specified
>
>  $ qemu-system-cris -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Kernel image must be specified
>
>  $ qemu-system-m68k -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Kernel image must be specified
>
>  $ qemu-system-mips -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait qemu: Could not load MIPS bios 'mips_bios.bin', and no -kernel argument was specified
>
>  $ qemu-system-mipsel -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  qemu: Could not load MIPS bios 'mipsel_bios.bin', and no -kernel argument was specified
>
>  $ qemu-system-or32 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Can't create serial device, empty char device
>
>  $ qemu-system-ppcemb -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Unable to find PowerPC CPU definition
>
>  $ qemu-system-sh4 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Initializing CPU
>  Allocating ROM
>  Allocating SDRAM 1
>  Allocating SDRAM 2
>  shix_init: load BIOS 'shix_bios.bin'
>  ret=-1
>  qemu: could not load SHIX bios 'shix_bios.bin'
>
>  $ qemu-system-sh4eb -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Initializing CPU
>  Allocating ROM
>  Allocating SDRAM 1
>  Allocating SDRAM 2
>  shix_init: load BIOS 'shix_bios.bin'
>  ret=-1
>  qemu: could not load SHIX bios 'shix_bios.bin'
>
>  $ qemu-system-unicore32 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  qemu-system-unicore32: /home/berrange/src/virt/qemu/hw/unicore32/../puv3.c:81: puv3_load_kernel: Assertion `kernel_filename != ((void *)0)' failed.
>  Aborted
>
>
> Can we "fix" them in some way to not require the kernel ?
>
> Do we have to go down the route of adding some sort of "-no-vm" flag to
> explicitly say we don't care about any of the VM setup parts ? This would
> be more like what my old patches did which mapped monitors commands
> onto '-query-XXXXX' command line args, bypassing VM setup, but still
> using normal QMP monitor interaction.
>
> Any other suggestions on how to reliably get a QMP monitor to a target,
> without any VM config. ?

-M none.  See the patch I just sent out.  I think we can include this in
1.2 as it poses no real risk.

Regards,

Anthony Liguori

>
> Daniel
> -- 
> |: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
> |: http://libvirt.org              -o-             http://virt-manager.org :|
> |: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
> |: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

[Qemu-devel] unicore32 fails assertion without -kernel (was: How to reliably start a bare QEMU target to query capabilities via QMP)

[Markus Armbruster]

"Daniel P. Berrange" <berrange@redhat.com> writes:

> I've been adapting libvirt to use to the various new QMP commands to
> query QEMU's capabilities, instead of the hated -help parsing. Obviously
> the critical part of this is being able to reliably start a bare QEMU
> process with no actual guest OS configured (no disks, no kernel, etc)
> and talk to its monitor. I hadn't anticipated problems since I only
> tested with x86_64 / i386 most of the time, which work well in this
> respect. Currently I am doing
>
>   # $QEMU_BINARY -S \
>          -no-user-config \
>          -nodefconfig \
>          -nodefaults \
>          -nographic \
>          -qmp stdio
>
>
> This works for about 50% of the QEMU targets:
[...]
>  but fails for the other 50% of targets:
[...]
> With the failing targets i see the following kinds of errors:
>
>
>  $ qemu-system-arm -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  Kernel image must be specified
[...]
>  $ qemu-system-unicore32 -S -nodefconfig -nodefaults -nographic -qmp unix:/tmp/foo,server,nowait
>  qemu-system-unicore32: /home/berrange/src/virt/qemu/hw/unicore32/../puv3.c:81: puv3_load_kernel: Assertion `kernel_filename != ((void *)0)' failed.
>  Aborted

Bug, cc'ing maintainer.  It should fail cleanly, like the test case
above.

A similar bug was recently fixed for armv7m (commit 01fd41ab).

[...]

Re: [Qemu-devel] How to reliably start a bare QEMU target to query capabilities via QMP

[Daniel P. Berrange]

On Wed, Aug 22, 2012 at 03:25:38PM -0500, Anthony Liguori wrote:
> "Daniel P. Berrange" <berrange@redhat.com> writes:
> 
> > I've been adapting libvirt to use to the various new QMP commands to
> > query QEMU's capabilities, instead of the hated -help parsing. Obviously
> > the critical part of this is being able to reliably start a bare QEMU
> > process with no actual guest OS configured (no disks, no kernel, etc)
> > and talk to its monitor. I hadn't anticipated problems since I only
> > tested with x86_64 / i386 most of the time, which work well in this
> > respect. Currently I am doing
> >
> >   # $QEMU_BINARY -S \
> >          -no-user-config \
> >          -nodefconfig \
> >          -nodefaults \
> >          -nographic \
> >          -qmp stdio
> >
> >
> > This works for about 50% of the QEMU targets:


> > Do we have to go down the route of adding some sort of "-no-vm" flag to
> > explicitly say we don't care about any of the VM setup parts ? This would
> > be more like what my old patches did which mapped monitors commands
> > onto '-query-XXXXX' command line args, bypassing VM setup, but still
> > using normal QMP monitor interaction.
> >
> > Any other suggestions on how to reliably get a QMP monitor to a target,
> > without any VM config. ?
> 
> -M none.  See the patch I just sent out.  I think we can include this in
> 1.2 as it poses no real risk.

Great, that's an nice elegant solution to the problem :-)


Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|