Re: [Xen-devel] [PATCH v2] tools/libxl: mark special pages as reserved in e820 map for PVH

[Juergen Gross <jgross@xxxxxxxx>]

On 21/11/17 12:27, Jan Beulich wrote:
>>>> On 21.11.17 at 12:06, <jgross@xxxxxxxx> wrote:
>> The "special pages" for PVH guests include the frames for console and
>> Xenstore ring buffers. Those have to be marked as "Reserved" in the
>> guest's E820 map, as otherwise conflicts might arise later e.g. when
>> hotplugging memory into the guest.
> 
> Afaict this differs from v1 only in no longer adding the extra entry
> for HVM. How does this address the concerns raised on v1 wrt spec
> compliance? v1 having caused problems with hvmloader should not
> have resulted in simply excluding HVM here. That's even more so
> because we mean HVM and PVH to converge in the long run - I'd
> expect that to mean that no clear type distinction would exist
> anymore on libxl.

The difference is for HVM the HVMloader is creating the additional
"Reserved" entry.

> If you want to reserve Xenstore ring buffer and console page,
> why don't you reserve just the two (provided of course they
> live outside of any [fake] PCI device BAR), which then ought to
> also be compatible with plain HVM?

For PVH the "mmio" area is starting with the LAPIC and extends up
to 4GB. And all of this area conflicts with the HVMloader:

(d11) HVM Loader
(d11) Detected Xen v4.10-unstable
(d11) Fail to setup memory map due to conflict on dynamic reserved
memory range.
(d11) *** HVMLoader bug at e820.c:52
(d11) *** HVMLoader crashed.

I guess this should be fixed, but I wanted the patch to be as small
as possible to minimze the risk for 4.10.


Juergen

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