Re: [Qemu-devel] [PATCH v7 41/42] Disable mlock around incoming postcopy

[zhanghailiang]

On 2015/8/6 22:55, Dr. David Alan Gilbert wrote:
> * Juan Quintela (address@hidden) wrote:
> > Amit Shah <address@hidden> wrote:
> > > On (Tue) 14 Jul 2015 [17:22:13], Juan Quintela wrote:
> > > > "Dr. David Alan Gilbert (git)" <address@hidden> wrote:
> > >
> > > > > +    if (enable_mlock) {
> > > > > +        if (os_mlock() < 0) {
> > > > > +            error_report("mlock: %s", strerror(errno));
> > > > > +            /*
> > > > > +             * It doesn't feel right to fail at this point, we have a valid
> > > > > +             * VM state.
> > > > > +             */
> > > >
> > > > realtime_init() exit in case of os_mlock() fails, so current code is:
> > >
> > > Yea, I was wondering the same - but then I thought: would the realtime
> > > case want a migration to happen at all?
> >
> > Then disable migration with realtime looks like saner.  But that
> > decission don't belong to this series.
>
> I added this patch because Zhanghailiang had reported trying to use it and it
> failing.
>
> Zhanghailiang: Do you have a use case for mlock=on and migration?

Yes, we usually configure mlock=on for VM that needs high performance, and we 
also support migration for these VMs.
It works well for pre-copy migration with this configuration.
IMHO, it is better to support this for post-copy migration too~

(Or maybe we could add dynamically mlock/munlock memory command for qemu,
and let management layer to decide what to do if they want to post-copy VM with 
mlocking memory ?).


> > > > - we start qemu with mlock requset
> > > > - we mlock memory
> > > > - we start postcopy
> > > > - we munlock memory
> > > > - we mlock memory
> > > >
> > > > I wmill really, really preffer having a check if memory is mlocked, and
> > > > it that case, just abort migration altogether.  Or better still, wait to
> > > > enable mlock *until* we have finished postcopy, no?
> > >
> > >                 Amit
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK
>
> .