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 13, 2016 at 01:08:57PM -0400, Boris Ostrovsky wrote:
> On 07/13/2016 11:22 AM, Julien Grall wrote:
> > Hello,
> >
> > On 12/07/2016 17:58, Boris Ostrovsky wrote:
> >> On 07/12/2016 12:10 PM, Julien Grall wrote:
> >>> On 12/07/2016 16:08, Boris Ostrovsky wrote:
> >>>> On 07/12/2016 10:57 AM, Shannon Zhao wrote:
> >>>>> On 2016年07月12日 22:50, Wei Liu wrote:
> >>>>>> On Tue, Jul 12, 2016 at 10:42:07PM +0800, Shannon Zhao wrote:
> >>>>>>>>>>>>>>>>>>>>>>>>>> Does it mean we would need to update the slack
> >>>>>>>>>>>>>>>>>>>>>>>>>> to take into account the ACPI
> >>>>>>>>>>>>>>>>>>>>>>>>>> blob?
> >>>>>>>>>>>>>>>>>> Yes, we need to take into account the ACPI blob.
> >>>>>>>>>>>>>>>>>> Probably not in the
> >>>>>>>>>>>>>>>>>> slack but directly in mam_memkb.
> >>>>>>>>>>>>>> Sorry, I'm not sure understand this. I found the
> >>>>>>>>>>>>>> b_info->max_memkb but
> >>>>>>>>>>>>>> didn't find the slack you said. And how to fix this? Update
> >>>>>>>>>>>>>> b_info->max_memkb or the slack?
> >>>>>>>>>> Can you calculate the size of your payload and add that to
> >>>>>>>>>> max_memkb?
> >>>>>>>>>>
> >>>>>>>> Yeah, but the size will be changed if we change the tables in the
> >>>>>>>> future
> >>>>>>>> and this also should consider x86, right?
> >>>>>> That could easily be solved by introducing a function to
> >>>>>> calculate the
> >>>>>> size, right?
> >>>>> Oh, I'm not familiar with this. Let's clarify on this. It can add the
> >>>>> size to max_memkb after generating the ACPI tables and before loading
> >>>>> the tables to guest space and it doesn't have to add the size at
> >>>>> libxl__domain_build_info_setdefault(), right?
> >>>>
> >>>> This was discussed before: ACPI tables are part of RAM whose size is
> >>>> specified by the config file (and is reflected in max_memkb I
> >>>> believe).
> >>>> It may not be presented to the guest as RAM (i.e. on x86 it is labeled
> >>>> by BIOS (or whoever) as a dedicated type in e820) but it still resides
> >>>> in DIMMs.
> >>>
> >>> I don't think this was the conclusion of the thread. IHMO, "maxmem" is
> >>> the amount of RAM a guest could effectively use.
> >>>
> >>> Whilst the ACPI tables will be in the DIMM from the host point of
> >>> view. From a guest point of view it will be a ROM.
> >>
> >> The config file specifies resources provided by the host. How the guest
> >> views those resources is not important, I think.
> >
> > This would need to be clarified. For instance special pages (Xenstore,
> > Console...) are RAM from the host point of view but not taken into
> > account in the "maxmem" provided by the user. For my understanding,
> > some kB of the slack is used for that.
> 
> 
> Are these pages part of guest's address space?
> 
> 
> >
> >>
> >>>
> >>> It will affect some others part of the guest if we don't increment the
> >>> "maxmem" requested by the user. For ARM the ACPI blob will be exposed
> >>> at a specific address that is outside of the guest RAM (see the guest
> >>> memory layout in public/arch-arm.h).
> >>>
> >>> We chose this solution over putting in the RAM because the ACPI tables
> >>> are not easily relocatable (compare to the device tree, initrd and
> >>> kernel) so we could not take advantage of superpage in both stage-2
> >>> (hypervisor) and stage-1 (kernel) page table.
> >>
> >> Maybe this is something ARM-specific then. For x86 we will want to keep
> >> maxmem unchanged.
> >
> > I don't think what I described in my previous mail is ARM-specific.
> > The pressure will be more important on the TLBs, if Xen does not use
> > superpage in the stage 2 page tables (i.e EPT for x86) no matter the
> > architecture.
> >
> > IHMO, this seems to be a bigger drawback compare to add few more
> > kilobytes to maxmem in the toolstack for the ACPI blob. You will loose
> > them when creating the intermediate page table in any case.
> 
> 
> Why not have the guest ask for more memory in the config file then?
> 
> (OK, I can see that they can't ask for a few KB since we have MB
> resolution but they can ask for an extra 1MB)
> 

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.

Wei.

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