Re: [Xen-devel] [PATCH 4/4] x86/mm: Remove force-invalidate loop

[Jan Beulich <jbeulich@xxxxxxxx>]

On 23.12.2019 17:43, George Dunlap wrote:
> The comment about the "force-invalidate" loop gives two reasons for
> its existence:
> 
> 1. Breaking circular "linear pagetable" references
> 
> 2. Cleaning up partially-validated pages.
> 
> The first reason been invalid since XSA-240: Since then it hasn't been
> possible to generate circular linear pagetable references.
> 
> The second reason has been invalid since long before that: Before
> calling relinquish_memory(), domain_relinquish_resources() calls
> vcpu_destroy_pagetables() on each vcpu; this will in turn call
> put_old_guest_table() on each vcpu.  The result should be that it's
> not possible to have top-level partially-devalidated pagetables by the
> time we get to relinquish_memory().

There's a subtle difference: Up in the enumeration you correctly
say "partially-validated pages". Down here you also correctly state
that put_old_guest_table() deals with partially-devalidated
pagetables. The latter indeed shouldn't be able to make it here,
but what about ones that have been partially validated (and hence
not put into ->arch.old_guest_table) while the domain was still
alive? (I think the code change is still correct, because
partially validated pages ought to be taken care of correctly, but
I'd prefer if the description matched the change.)

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel