Re: [RFC PATCH] iothread: add set_iothread_poll_* commands

[Dr. David Alan Gilbert]

* Zhenyu Ye (address@hidden) wrote:
> 
> 
> On 2019/10/24 21:56, Dr. David Alan Gilbert wrote:
> > * Zhenyu Ye (address@hidden) wrote:
> >>
> >>
> >> On 2019/10/23 23:19, Stefan Hajnoczi wrote:
> >>> On Tue, Oct 22, 2019 at 04:12:03PM +0800, yezhenyu (A) wrote:
> >>>> Since qemu2.9, QEMU added three AioContext poll parameters to struct
> >>>> IOThread: poll_max_ns, poll_grow and poll_shrink. These properties are
> >>>> used to control iothread polling time.
> >>>>
> >>>> However, there isn't properly hmp commands to adjust them when the VM is
> >>>> alive. It's useful to adjust them online when observing the impact of
> >>>> different property value on performance.
> >>>>
> >>>> This patch add three hmp commands to adjust iothread poll-* properties
> >>>> for special iothread:
> >>>>
> >>>> set_iothread_poll_max_ns: set the maximum polling time in ns;
> >>>> set_iothread_poll_grow: set how many ns will be added to polling time;
> >>>> set_iothread_poll_shrink: set how many ns will be removed from polling
> >>>> time.
> >>>>
> >>>> Signed-off-by: Zhenyu Ye <address@hidden>
> >>>> ---
> >>>> hmp-commands.hx | 42 ++++++++++++++++++++
> >>>> hmp.c | 30 +++++++++++++++
> >>>> hmp.h | 3 ++
> >>>> include/sysemu/iothread.h | 6 +++
> >>>> iothread.c | 80 +++++++++++++++++++++++++++++++++++++++
> >>>> qapi/misc.json | 23 +++++++++++
> >>>> 6 files changed, 184 insertions(+)
> >>>
> >>> poll-max-ns, poll-grow, poll-shrink are properties of IOThread objects.
> >>> They can already be modified at runtime using:
> >>>
> >>>   $ qemu -object iothread,id=iothread1
> >>>   (qemu) qom-set /objects/iothread1 poll-max-ns 100000
> >>>
> >>> I think there is no need for a patch.
> >>>
> >>> Stefan
> >>>
> >>
> >> Thanks for your review. I have considered using the `qom-set` command to 
> >> modify
> >> IOThread object's properties, however, this command is not friendly to 
> >> primary
> >> users. The help info for this command is only:
> >>
> >>     qom-set path property value -- set QOM property
> >>
> >> It's almost impossible to get the correct `path` parameter for primary 
> >> user.
> > 
> > Is this just a matter of documenting how to do it?
> > 
> > It sounds like there's no need for a new QMP command though;  if you
> > want an easier HMP command I'd probably still take it (because HMP is ok
> > at having things for convenience) - but not if it turns out that just
> > adding a paragraph of documentation is enough.
> > 
> > Dave
> > 
> 
> I will show the differences in QMP and HMP:
> If I want to set iothread1.poll-max-ns=1000 and iothread1.poll-grow=2:
> 
> Without this patch:
> QMP command:
> 
>     qom-set /objects/iothread1 poll-max-ns 1000
>     qom-set /objects/iothread1 poll-grow 2
> 
> HMP command:
> 
>     { "execute": "qom-set", "arguments": { "path": "/objects/iothread1",
>                                            "property": "poll-max-ns", 
> "value": 1000 } }
>     { "execute": "qom-set", "arguments": { "path": "/objects/iothread1",
>                                            "property": "poll-grow", "value": 
> 2} }
> 
> with this patch:
> QMP command:
> 
>     iothread_set_parameter iothread1 max-ns 1000
>     iothread_set_parameter iothread1 grow 2
> 
> HMP command:
> 
>     { "execute": "set-iothread-poll-params", "arguments': { "iothread-id": 
> "iothread1",
>                                                             "max-ns": 1000, 
> "grow": 2 } }
> 
> 
> I think the inconvenience of qom-set is how to get the correct `path` 
> parameter.
> Anyway, I will consider your advice.

So it depends how obvious the path is;  if it's just   /objects/
followed by whatever you used with id=  when you created the iothread
then I think it's easy - we just need to update the docs.
Is there a case where it's harder to know?

Dave

> 
> >> This patch provides a more convenient and easy-use hmp&qmp interface to 
> >> modify
> >> these IOThread properties. I think this patch still has a little value.
> >>
> >> And I can implement this patch compactly by reusing your code.
> >>
> >> Waiting for your reply.
> >>
> > --
> > Dr. David Alan Gilbert / address@hidden / Manchester, UK
> > 
> > 
> > .
> > 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK