Re: [Qemu-devel] [RFC PATCH] vl: only display available accelerators

[Eduardo Habkost]

On Wed, Nov 08, 2017 at 05:06:29PM +0000, Daniel P. Berrange wrote:
> On Wed, Nov 08, 2017 at 02:59:05PM -0200, Eduardo Habkost wrote:
> > On Wed, Nov 08, 2017 at 01:21:33PM -0300, Philippe Mathieu-Daudé wrote:
> > > On 11/08/2017 10:28 AM, Daniel P. Berrange wrote:
> > > > On Mon, Oct 30, 2017 at 09:20:29AM +0100, Eduardo Habkost wrote:
> > > >> On Mon, Oct 30, 2017 at 01:00:56AM -0300, Philippe Mathieu-Daudé wrote:
> > > >>> examples configuring with '--enable-kvm --disable-tcg'
> > > >>>
> > > >>> - before
> > > >>>
> > > >>>   $ qemu-system-x86_64 -accel help
> > > >>>   Possible accelerators: kvm, xen, hax, tcg
> > > >>>
> > > >>>   $ qemu-system-x86_64 -accel tcg
> > > >>>   qemu-system-x86_64: -machine accel=tcg: No accelerator found
> > > >>>
> > > >>>   # qemu-system-x86_64 -accel hax
> > > >>>   qemu-system-x86_64: -machine accel=hax: No accelerator found
> > > >>>
> > > >>>   # qemu-system-x86_64 -accel xen
> > > >>>   xencall: error: Could not obtain handle on privileged command 
> > > >>> interface: No such file or directory
> > > >>>   xen be core: xen be core: can't open xen interface
> > > >>>   can't open xen interface
> > > >>>   qemu-system-x86_64: failed to initialize Xen: Operation not 
> > > >>> permitted
> > > >>>
> > > >>> - after
> > > >>>
> > > >>>   $ qemu-system-x86_64 -accel help
> > > >>>   Possible accelerators: kvm
> > > >>>
> > > >>> Signed-off-by: Philippe Mathieu-Daudé <address@hidden>
> > > >>> ---
> > > >>> RFC because:
> > > >>>     - I don't think this is the nicest way, too much #ifdef'fery in 
> > > >>> main()
> > > >>
> > > >> I suggest using object_class_get_list(TYPE_ACCEL, false).
> > > > 
> > > > And check the result of the available() method on the returned classes
> > > > too, to filter the results.
> > > 
> > > Good idea! I'll use that.
> > 
> > It looks like QTest is the only accelerator that implements
> > ->available(), and its return value is a build-time constant that
> > depends only on CONFIG_POSIX.
> > 
> > I wonder why we don't simply avoid compiling the qtest class if
> > CONFIG_POSIX is unset, making the ->available() method
> > unnecessary.
> 
> Yeah that does seem simpler, though I'm surprised that Xen does not
> implement the available method. Xen is an accel I'd expect to see
> compiled into an x86 build, but is only available if the host is
> actually booted under a Xen hypervisor.  Likewise shouldn't kvm
> only report itself as available if the /dev/kvm actually exists.
> But maybe that's not the kind of semantics code using available()
> expects ?

Currently the only caller of ->available() calls ->init_machine()
immediately afterwards, so for the current code it doesn't matter
if the check is inside ->available() or ->init_machine().

That said, I'm not sure we should look for /dev/kvm or check for
the Xen hypervisor when handling "-accel help".  I expect help
text to tell the user what the QEMU binary supports, not what the
current host supports.

-- 
Eduardo