Re: [Xen-devel] [PATCH v2 2/2] xen/balloon: Enforce various limits on target

[Ian Campbell <Ian.Campbell@citrix.com>]

On Fri, 2013-05-03 at 16:47 +0100, Daniel Kiper wrote:
> > > > > Do not forget that guest may change target itself.
> > > >
> > > > Yes it can, and that can fail either due to maxmem or due to ENOMEM, and
> > > > the kernel needs prepared to deal with that when it happens.
> > >
> > > Sure but why we would like to fail in endless loop if maxmem case
> > > could be easliy detected by checking XENMEM_maximum_reservation?
> >
> > That endless loop is deliberate. When a target is set the balloon driver
> > is supposed to try to reach it and if it fails at any given moment it is
> > supposed to try again. This all relates to the changes made in
> > bc2c0303226e.
> >
> > Now you could argue that this case is subtly different from the ENOMEM
> > case which was the primary focus of that commit but have you thought
> > about the behaviour you want in the case where maximum_reservation is
> > subsequently raised? IMHO there's no reason why the balloon driver
> > shouldn't then automatically make further progress towards the target.
> 
> OK, now it makes sens. Do we assume the same behavior for dom0?
> Could we change maximum_reservation for dom0 using xl?

I don't think there's any reason to special case dom0 here.

> > If the infinite loop bothers you then perhaps an exponential backoff in
> > the frequency of attempts would be a suitable middle ground?
> 
> Relevant patches made by me are merged some time ago.

Great!

> > > > > Additionally, we would like to introduce xm compatibility
> > > > > mode which is a bit different then xl normal behavior.
> > > >
> > > > When then you really don't want to be baking specifics of the current
> > > > model into the kernel, do you.
> > >
> > > Hmmm... Little misunderstanding. As I stated a few times I do not
> > > want bake any libxl or Xen stuff into Linux Kernel (including
> > > LIBXL_MAXMEM_CONSTANT). I just want to check limits which I think
> > > make sense in this case.
> >
> > Sorry, I never noticed you saying that. Where was it?
> 
> Here http://lists.xen.org/archives/html/xen-devel/2013-04/msg03259.html
> I stated that I do not like this constant. I explained why this was done
> in that way.

Sorry, I read that mail as arguing that it must be done this way ("this
is a must").

>  Later I found relevant commit which introduced it and asked
> authors about it. I think that shows that I am not happy with
> LIBXL_MAXMEM_CONSTANT and I am looking for good solution for this problem.

Ian.