Re: [Qemu-devel] [PATCH v3 4/6] qemu-option: support +foo/-foo command line agruments

[Igor Mammedov <imammedo@redhat.com>]

On Wed, 13 Nov 2013 13:07:26 +1100
Alexey Kardashevskiy <aik@ozlabs.ru> wrote:

> On 11/13/2013 12:11 AM, Igor Mammedov wrote:
> > On Tue, 12 Nov 2013 23:39:27 +1100
> > Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> > 
> >> On 12.11.2013 20:58, Igor Mammedov wrote:
> >>> On Tue, 12 Nov 2013 10:49:58 +1100
> >>> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> >>>
> >>>> On 11/12/2013 01:25 AM, Igor Mammedov wrote:
> >>>>> On Mon, 11 Nov 2013 13:41:05 +0100
> >>>>> Andreas Färber <afaerber@suse.de> wrote:
> >>>>>
> >>>>>> Am 11.11.2013 08:44, schrieb Alexey Kardashevskiy:
> >>>>>>> This converts +foo/-foo to "foo=on"/"foo=off" respectively when
> >>>>>>> QEMU parser is used for the command line options.
> >>>>>>>
> >>>>>>> "-cpu" parsers in x86 and other architectures should be unaffected
> >>>>>>> by this change.
> >>>>>>>
> >>>>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>>>>> ---
> >>>>>>>  util/qemu-option.c | 6 ++++++
> >>>>>>>  1 file changed, 6 insertions(+)
> >>>>>>>
> >>>>>>> diff --git a/util/qemu-option.c b/util/qemu-option.c
> >>>>>>> index efcb5dc..6c8667c 100644
> >>>>>>> --- a/util/qemu-option.c
> >>>>>>> +++ b/util/qemu-option.c
> >>>>>>> @@ -890,6 +890,12 @@ static int opts_do_parse(QemuOpts *opts, const char *params,
> >>>>>>>                  if (strncmp(option, "no", 2) == 0) {
> >>>>>>>                      memmove(option, option+2, strlen(option+2)+1);
> >>>>>>>                      pstrcpy(value, sizeof(value), "off");
> >>>>>>> +                } else if (strncmp(option, "-", 1) == 0) {
> >>>>>>> +                    memmove(option, option+1, strlen(option+1)+1);
> >>>>>>> +                    pstrcpy(value, sizeof(value), "off");
> >>>>>>> +                } else if (strncmp(option, "+", 1) == 0) {
> >>>>>>> +                    memmove(option, option+1, strlen(option+1)+1);
> >>>>>>> +                    pstrcpy(value, sizeof(value), "on");
> >>>>>>>                  } else {
> >>>>>>>                      pstrcpy(value, sizeof(value), "on");
> >>>>>>>                  }
> >>>>>>
> >>>>>> This looks like an interesting idea! However this is much too big a
> >>>>>> change to just CC ppc folks on...
> >>>>>>
> >>>>>> Jan, I wonder if this might break slirp's hostfwd option?
> >>>>>>
> >>>>>> Not sure what other options potentially starting with '-' might be
> >>>>>> affected. Test cases would be a helpful way of demonstrating that this
> >>>>>> change does not have undesired side effects.
> >>>>> on x86 there is several value fixups for compatibility reason and a manual
> >>>>> value parsing in cpu_x86_parse_featurestr(), so above won't just work there.
> >>>>
> >>>>
> >>>> What particular x86 CPU option cannot be handled the way as PPC's "VSX" is
> >>>> handled two patches below? As I see, even static properties will work there
> >>>> fine.
> >>> There is legacy code that is kept for CLI compatibility reasons.
> >>> Please, look at following features in cpu_x86_parse_featurestr():
> >>>   xlevel, tsc-freq hv-spinlocks
> >>
> >> Ok, I do not know for sure if static properties support setters/getters
> >> (they do not if I remember correct) but what does prevent these x86
> >> properties from being _dynamic_?
> > nothing, except of:
> >  * it's better to keep CPU device model clean from legacy hacks so that legacy
> >    silent fixups of invalid values won't be available via other interfaces
> >    except of CLI. That will force users to use correct property names/values
> >    and not break old users that use legacy CLI options.
> >
> >>> the rest feature flags on x86 should be handled just fine by your patch,
> >>> once x86properties series is applied. 
> >>>
> >>> that's why we are talking about parser hook that could be overridden
> >>> by target if necessary.
> >>
> >> This part confuses me the most. I thought I added the hook and I did not
> >> change other than PPC archs so my patches should have gone quite easily
> >> to upstream but instead I was told (I think I was but I could
> >> misunderstand) that other folks may be unhappy that my stuff does not
> >> support +foo/-foo (which could be added later).
> >>
> >> Could you please point me to the x86properties patch(es) which everybody
> >> is waiting for? Thanks!
> > latest is available at
> > https://github.com/imammedo/qemu/tree/x86-cpu-properties.v10.1
> > which basically is a rebase with fixed conflicts of v9
> > http://comments.gmane.org/gmane.comp.emulators.qemu/222284
> 
> 
> Wow. This explains a lot. Thanks. Is there any plan to use QemuOpts for all
> of this, instead of cpu_x86_parse_featurestr()?
the plan was to keep, +/-/fixups as legacy in target specific code (x86,
sparc) not polluting the rest targets. For not affected targets use only
foo=val notation in CLI/monitor.

So providing a generic parser of cpu_model string for most targets and having
a hook override with custom parser on x86,sparc would be one of simplest
solutions.

> 
> 
> -- 
> Alexey


-- 
Regards,
  Igor