Re: [Qemu-devel] host physical address width issues/questions for x86_64

[Prasad Singamsetty]

On 10/13/2017 10:01 AM, Dr. David Alan Gilbert wrote:
> * Prasad Singamsetty (address@hidden) wrote:
> > Hi,
> >
> > I am new to the alias. I have some questions on this subject
> > and seek some clarifications from the experts in the team.
> > I ran into a couple of issues when I tried with large configuration
> > ( >= 1TB memory, > 255 CPUs) for x86_64 guest machine.
> >
> > 1. QEMU uses the default value of 40 (TCG_PHYS_ADDR_BITS) for address
> >     width if user has not specified phys-bits or host-phys-bits=true
> >     property. The default value is obviously not sufficient and
> >     causing guest kernel to crash if configured with >= 1TB
> >     memory. Depending on the linux kernel version in the guest the
> >     panic was in different code paths. The workaround is for the
> >     user to specify the phys-bits property or set the property
> >     host-phys-bits=true.
> >
> >     QUESTIONS:
> >     1) Could we change the default value to same as the host physcial
> >        address for x86_64 machines?  Are there any side effects on this?
> That's what we do in the RH downstream packages.
>
> If you did that you wouldn't want to break existing machine-types,
> so you'd have to tie it to a new machine type.
OK.

> There's some fun with MTRRs that have bits set based on the address
> size, and if you migrate between hosts with different physical address sizes; 
> e.g. between
> a non-Xeon (or I think a Xeon-E3) and the bigger boxes you have
> to be careful.  See fcc35e7 and commits around that;  tbh I can't
> remember the details.
Right. The problem with migration between hosts is still there.

> >     2) Adding a check to fail to boot the guest if phys-bits is not
> >        sufficient for the specified maxmem or if it is more than
> >        the host phys bits value. Do you have any objections if I
> >        add a patch for this?
> It's a little more complicated, but good in principal.  You need
> to take account of the allocated address space for hotplug
> and I think the PCI address space;  I can't remember if we
> ever figured out a good way of finding that out.
> I think it might also depend if you're on SeaBIOS or OVMF
> about what they're defaults are for things like where PCI
> gets allocated.
Thanks for the suggestions. I will check with OVMF also.

> > 2. host_address_width in DMAR table structure
> >
> >     In this case, the default value is set to 39
> >     (VTD_HOST_ADDRESS_WIDTH - 1). With interrupt remapping
> >     enabled for the intel iommu and the guest is configured
> >     with > 255 cpus and >= 1TB memory, the guest kernel hangs
> >     during boot up. This need to be fixed.
> >
> >     QUESTION:
> >     The question here again is can we fix this to use the
> >     real address width from the host as the default?
> I don't know DMAR stuff; chatting to Alex (cc'd) it does sound
> like that's an ommission that should be fixed.
Thanks,
--Prasad

> > Please let me know if you have some suggestions in fixing these
> > two problem cases for supporting large config guests. Also, please
> > let me know if there are any other known limitations in the current
> > implementation.
> Dave
>
> > Thanks.
> > --Prasad
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK