On Wed, Sep 08, 2021 at 04:46:19PM +0100, Alexandru Elisei wrote:
> Hi Drew,
>
> On 9/8/21 4:09 PM, Andrew Jones wrote:
> > On Wed, Sep 08, 2021 at 03:33:19PM +0100, Alexandru Elisei wrote:
> > ...
> >>>> +fixup_kvmtool_opts()
> >>>> +{
> >>>> + local opts=$1
> >>>> + local groups=$2
> >>>> + local gic
> >>>> + local gic_version
> >>>> +
> >>>> + if find_word "pmu" $groups; then
> >>>> + opts+=" --pmu"
> >>>> + fi
> >>>> +
> >>>> + if find_word "its" $groups; then
> >>>> + gic_version=3
> >>>> + gic="gicv3-its"
> >>>> + elif [[ "$opts" =~ -machine\ *gic-version=(2|3) ]]; then
> >>>> + gic_version="${BASH_REMATCH[1]}"
> >>>> + gic="gicv$gic_version"
> >>>> + fi
> >>>> +
> >>>> + if [ -n "$gic" ]; then
> >>>> + opts=${opts/-machine gic-version=$gic_version/}
> >>>> + opts+=" --irqchip=$gic"
> >>>> + fi
> >>>> +
> >>>> + opts=${opts/-append/--params}
> >>>> +
> >>>> + echo "$opts"
> >>>> +}
> >>> Hmm, I don't think we want to write a QEMU parameter translator for
> >>> all other VMMs, and all other VMM architectures, that we want to
> >>> support. I think we should add new "extra_params" variables to the
> >>> unittest configuration instead, e.g. "kvmtool_params", where the
> >>> extra parameters can be listed correctly and explicitly. While at
> >>> it, I would create an alias for "extra_params", which would be
> >>> "qemu_params" allowing unittests that support more than one VMM
> >>> to clearly show what's what.
> >> I agree, this is a much better idea than a parameter translator. Using a dedicated
> >> variable in unittests.cfg will make it easier for new tests to get support for all
> >> VMMs (for example, writing a list of parameters in unittests.cfg should be easier
> >> than digging through the scripts to figure exactly how and where to add a
> >> translation for a new parameter), and it allow us to express parameters for other
> >> VMMs which don't have a direct correspondent in qemu.
> >>
> >> By creating an alias, do you mean replacing extra_params with qemu_params in
> >> arm/unittests.cfg? Or something else?
> > Probably something like this
> >
> > diff --git a/scripts/common.bash b/scripts/common.bash
> > index 7b983f7d6dd6..e5119ff216e5 100644
> > --- a/scripts/common.bash
> > +++ b/scripts/common.bash
> > @@ -37,7 +37,12 @@ function for_each_unittest()
> > elif [[ $line =~ ^smp\ *=\ *(.*)$ ]]; then
> > smp=${BASH_REMATCH[1]}
> > elif [[ $line =~ ^extra_params\ *=\ *(.*)$ ]]; then
> > - opts=${BASH_REMATCH[1]}
> > + elif [[ $line =~ ^extra_params\ *=\ *(.*)$ ]]; then
> > + qemu_opts=${BASH_REMATCH[1]}
> > + elif [[ $line =~ ^qemu_params\ *=\ *(.*)$ ]]; then
> > + qemu_opts=${BASH_REMATCH[1]}
> > + elif [[ $line =~ ^kvmtool_params\ *=\ *(.*)$ ]]; then
> > + kvmtool_opts=${BASH_REMATCH[1]}
> > elif [[ $line =~ ^groups\ *=\ *(.*)$ ]]; then
> > groups=${BASH_REMATCH[1]}
> > elif [[ $line =~ ^arch\ *=\ *(.*)$ ]]; then
> >
> > and all other changes needed to support the s/opts/qemu_opts/ change
> > should work. Also, an addition to the unittests.cfg documentation.
>
> Got it, replace extra_opts with qemu_opts in the scripts.
>
> Yes, the documentation for unittests.cfg (at the top of the file) should
> definitely be updated to document the new configuration option, kvmtool_params.
>
> >
> > The above diff doesn't consider that a unittests.cfg file could have
> > both an 'extra_params' and a 'qemu_params' field, but I'm not sure
> > we care about that. Users should read the documentation and we
> > should review changes to the committed unittests.cfg files to avoid
> > that.
>
> What do you feel about renaming extra_params -> qemu_params in unittests.cfg?
Yes, that's what I would expect the patch to do.
> I'm
> thinking it would make the usage clearer, improve consistency (we would have
> qemu_params and kvmtool_params, instead of extra_params and kvmtool_params), and
> remove any confusions regarding when they are used (I can see someone thinking
> that extra_params are used all the time, and are appended to kvmtool_params when
> --target=kvmtool). On the other hand, this could be problematic for people using
> out-of-tree scripts that parse the unittest.cfg file for whatever reason (are
> there people that do that?).
I'm not as worried about that as about people using out-of-tree
unittests.cfg files that will break when the 'extra_params' field
disappears. That's why I suggested to make 'extra_params' an alias.
Thanks,
drew
>
> Thanks,
>
> Alex
>
> >
> > Thanks,
> > drew
> >
>
