Re: [Xen-devel] [PATCH raisin 4/4] Update to 4.7, update qemu and qemu_traditional recipes

[George Dunlap <george.dunlap@xxxxxxxxxx>]

On 14/06/16 14:53, Anthony PERARD wrote:
> On Tue, Jun 14, 2016 at 11:34:43AM +0100, George Dunlap wrote:
>> On 14/06/16 11:31, Stefano Stabellini wrote:
>>> On Tue, 14 Jun 2016, George Dunlap wrote:
>>>> On 14/06/16 11:08, Stefano Stabellini wrote:
>>>>> On Tue, 14 Jun 2016, George Dunlap wrote:
>>>>>> On 14/06/16 10:46, Stefano Stabellini wrote:
>>>>>>> On Mon, 13 Jun 2016, George Dunlap wrote:
>>>>>>>> Add a 4.7 config file, make it the default.
>>>>>>>>
>>>>>>>> Also update the qemu and qemu_traditional recipies after Ian Cambell's
>>>>>>>> work to split off separate libraries.
>>>>>>>>
>>>>>>>> Signed-off-by: George Dunlap <george.dunlap@xxxxxxxxxx>
>>>>>>>> ---
>>>>>>>> CC: Stefano Stabellini <sstabellini@xxxxxxxxxx>
>>>>>>>> ---
>>>>>>>>  components/qemu             | 21 +++++++++++++--------
>>>>>>>>  components/qemu_traditional |  2 +-
>>>>>>>>  configs/config-4.7          |  8 ++++++++
>>>>>>>>  defconfig                   |  2 +-
>>>>>>>>  4 files changed, 23 insertions(+), 10 deletions(-)
>>>>>>>>
>>>>>>>> diff --git a/components/qemu b/components/qemu
>>>>>>>> index e0d92a5..8624b50 100644
>>>>>>>> --- a/components/qemu
>>>>>>>> +++ b/components/qemu
>>>>>>>> @@ -23,15 +23,20 @@ function qemu_build() {
>>>>>>>>      cd "$BASEDIR"
>>>>>>>>      git-checkout $QEMU_URL $QEMU_REVISION qemu-dir
>>>>>>>>      cd qemu-dir
>>>>>>>> -    ./configure --enable-xen --target-list=i386-softmmu 
>>>>>>>> --prefix=$PREFIX \
>>>>>>>> -        --extra-cflags="-I$INST_DIR/$PREFIX/include" \
>>>>>>>> -        --extra-ldflags="-L$INST_DIR/$PREFIX/lib 
>>>>>>>> -Wl,-rpath-link=$INST_DIR/$PREFIX/lib \
>>>>>>>> +
>>>>>>>> +    ./configure --enable-xen --target-list=i386-softmmu \
>>>>>>>> +              --prefix=$PREFIX \
>>>>>>>> +              --extra-cflags="-DXC_WANT_COMPAT_EVTCHN_API=1 \
>>>>>>>> +              -DXC_WANT_COMPAT_GNTTAB_API=1 \
>>>>>>>> +              -DXC_WANT_COMPAT_MAP_FOREIGN_API=1 \
>>>>>>>> +                -I$INST_DIR/$PREFIX/include" \
>>>>>>>> +              --extra-ldflags="-L$INST_DIR/$PREFIX/lib 
>>>>>>>> -Wl,-rpath-link=$INST_DIR/$PREFIX/lib \
>>>>>>>>                           -L$INST_DIR/$PREFIX/lib64 
>>>>>>>> -Wl,-rpath-link=$INST_DIR/$PREFIX/lib64" \
>>>>>>>> -        --disable-kvm \
>>>>>>>> -        --disable-docs \
>>>>>>>> -        --bindir=$PREFIX/lib/xen/bin \
>>>>>>>> -        --datadir=$PREFIX/share/qemu-xen \
>>>>>>>> -        --disable-guest-agent
>>>>>>>> +              --bindir=$PREFIX/lib/xen/bin \
>>>>>>>> +              --datadir=$PREFIX/share/qemu-xen \
>>>>>>>> +              --disable-kvm \
>>>>>>>> +              --disable-docs \
>>>>>>>> +              --disable-guest-agent
>>>>>>>
>>>>>>> This is adding XC_WANT_COMPAT_EVTCHN_API=1, etc, unconditionally. If we
>>>>>>> make this change, will raisin still be able to buil an older qemu
>>>>>>> against an older xen?
>>>>>>
>>>>>> I've tested this with XEN_RELEASE set to 4.5 and 4.6, and it builds just
>>>>>> fine.  The #define is just ignored, since neither Xen nor qemu in the
>>>>>> 4.5 and 4.6 branches know anything about it.
>>>>>>
>>>>>> If part of our goal is to be a repository of the canonical way to build
>>>>>> things, though, perhaps it would be better to have the extra flags be
>>>>>> conditional on the value of XEN_RELEASE.
>>>>>
>>>>> If we are sure that it's ignored than it might be benign. Maybe we
>>>>> should just add an in-code comment to explain why we added it though.
>>>>
>>>> Well really, I think that the detection should be put into the qemu
>>>> configure scripts -- i.e., it should detect that there's the new API and
>>>> add the appropriate #defines itself.  But that's the sort of change I
>>>> don't think there will be a lot of enthusiasm for. :-)
>>>
>>> Actually there is already code in QEMU configure script for the
>>> detection. In fact I was wondering if explicitly passing XC_WANT_* is
>>> actually necessary. I don't think it is from QEMU POV.
>>
>> That code might be in qemu-upstream, but it's not in the qemu-xen trees;
>> if you revert this bit of the patch and try to build with raisin it will
>> fail.  That's why I even bothered to add these in in the first place;
>> and why IanC added these runes to tools/Makefile.
>>
>> Anthony, can you comment more authoritatively here on what's going on?
> 
> Yes, easy, xen-4.7 introduce a new API but qemu-xen-4.7 does not know
> about it, so qemu-xen will fail to compile. But there are some magic
> flags that are use to provide an compatible API and can qemu-xen
> compile.
> 
> QEMU have been taught about the new API and will use it, without the
> magic flags. QEMU will actually ignore the magic flags with a bunch of
> #undef.

So wait, I'm confused.  Isn't qemu-xen supposed to be upstream qemu plus
a handful of patches on top?  Or has there not been a QEMU release since
the time the new API?  Why on earth are we shipping in 4.7 a qemu that
can't use the new API we define in 4.7?

>> Would it be possible for 4.7.1 to backport the configure auto-detection
>> stuff?
> 
> I think that would be possible. But backporting the use of the new api
> will probably be harder. I guess we could just have qemu-xen define the
> magic flags itself instead of relying on them been given via a configure
> option.

Well at this point it's a bit late to import support for the new API
into qemu-xen-4.7; but it would be good for 4.7.1 to have qemu-xen's
configure set the necessary flags rather than bodging them in via
"--extra-cflags" in the Xen makefile.

 -George


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel