Re: [Xen-devel] [PATCH RFC] x86/HVM: fix various aspects of x2APIC emulation

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 09.09.14 at 18:56, <andrew.cooper3@xxxxxxxxxx> wrote:
> In the past, it has been suggested that doing compatibility fixing like
> this in the toolstack is far better than in Xen.  A toolstack is in a
> better position to know that a migration is occurring from an older
> version of xen to a newer version, and across this difference, LDR needs
> fixing up in the lapic state.  Furthermore, an ability like this can be
> used in reverse to safely migrate backwards, a requested feature of
> XenServer incase a partial upgrade needs to be stopped or undone.

Thing is - with or without migration, clustered mode is completely
broken with the incorrect setting up of LDR. Detecting and correcting
the bad LDR setting is safe (if the guest worked before, it must not
have run in clustered mode, i.e. doesn't care about LDR changing
under its feet). And seeing the bad LDR setting strictly implies that
the ID register wasn't set properly either. And again - so far it was
also always said that migration _to_ an older version of Xen is not a
supported scenario.

> Is it possible to split this patch between the safe cleanup and the
> unsafe cleanup?

Some of the pieces could be broken out, yes (including stuff I just
added, like forcing #GP upon invalid APIC base MSR transitions).
But already the widening of the "dest" function parameters would
better not be separated from the ID correction.

Jan


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