Re: [Xen-devel] [PATCH v2 16/17] libxc/xc_dom_arm: Copy ACPI tables to guest space

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Wed, Jul 20, 2016 at 02:52:05PM +0800, Shannon Zhao wrote:
> 
> 
> On 2016/7/19 18:38, Wei Liu wrote:
> > On Fri, Jul 15, 2016 at 05:39:32PM +0800, Shannon Zhao wrote:
> > [...]
> >>> > > 
> >>> > > It would be trivial to have another option in xl.cfg to allow MB
> >>> > > granularity. But I don't think that's a good idea. Asking for more
> >>> > > memory when you don't really know how much is enough is not very 
> >>> > > useful.
> >>> > > If an admin can know how much is needed, surely the library can be
> >>> > > taught to obtain that knowledge, too.
> >>> > > 
> >>> > > We need to decide which model we should go with. And, if we decide to
> >>> > > diverge, document the difference between x86 and ARM model.
> >>> > > 
> >> > Hi Wei,
> >> > 
> >> > Do you decide how to add the size of ACPI blob to max_memkb?
> >> > 
> > AFAICT ARM and x86 maintainers hold different opinions on how memory
> > should be accounted.
> > 
> > I would like to have a unified memory accounting model. But if we can't
> > have that at the moment, I'm fine with divergence, but please document
> > it somewhere (comment near code snippet, in header, or a file under docs
> > etc). And the amount added to max_memkb needs to be properly calculated,
> > not some magic number, so that we have a chance in the future to
> > confidently change how we do thing.
> If it's only allowed to add the size to max_memkb in
> libxl__domain_build_info_setdefault(), it only can use a const number
> since the tables are not generted and we don;t know the size. But the
> const number could be chosen properly since we could know the maximum
> ACPI tables size of current ARM approach.
> 
> But maybe in the future, if we add some new ACPI tables, it should
> increase the size as well. So I think this should be documented.
> 
> As I asked before, is it ok to add the size to max_memkb after
> generating the ACPI tables and before loading the tables to guest space?
> 

Yes. I don't think shoehorning everything into setdefault is a good
idea.

I think libxl_arm.c:libxl__arch_domain_create would be the right place
to do it.

I am thinking about calling xc_domain_setmaxmem there, but not adding a
number to d_config->b_info.max_memkb. Adding that to ->max_memkb would
be wrong because the bigger ->max_memkb will be recored and the same
algorithm will be applied every time you migrate your guest, so the
max_memkb will grow bigger and bigger.

Given the different approach taken by ARM and x86, maybe we need to also
record the size of acpi blobs somewhere in xenstore (also needs to be
documented) so that subsequent libxl_domain_setmaxmem can extract that
number again.

Please wait a bit for Stefano and Julien to comment before you do work.

I know, memory accounting is a bit messy. I don't claim that I figure
every last detail out. If I'm not clear enough or ignores some basic
facts do let me know.

Wei.

> Thanks,
> -- 
> Shannon
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel