Re: [PATCH 2/2] x86/svm: Add checks for nested HW features

From: Brian Woods <brian.woods@amd.com>
To: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>,
	Brian Woods <brian.woods@amd.com>,
	Jan Beulich <jbeulich@suse.com>,
	Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>,
	xen-devel@lists.xen.org
Subject: Re: [PATCH 2/2] x86/svm: Add checks for nested HW features
Date: Fri, 5 Jan 2018 09:27:30 -0600	[thread overview]
Message-ID: <20180105152730.GA11036@amd.com> (raw)
In-Reply-To: <8eb9d4df-e911-c213-4c32-dd20a54d0728@citrix.com>

On Fri, Dec 22, 2017 at 03:15:48PM +0000, Andrew Cooper wrote:
> Unfortunately, nestedhvm_enabled() is guaranteed to be false at the
> point that construct_vmcb() is called (due the order in which
> information appears while constructing the VM), which means we will
> never enable these optimisations.
> 
> Combined with the observation of EFER in the pipeline, the logic to
> enable/disable these optimisations needs to be in
> svm_update_guest_efer(), and need to trigger when EFER.SVME changes.
> 
> ~Andrew

Sorry for the late reply.  I tired working this before vacation but it
turned out to be a little bit longer than that... I have a set of
patches that _should_ work, but there are other issues.  Turns out there
are interrupt issues with nestted SVM HVM and I'm trying to hunt those
down and fix them so I can properly test the patches I've done.  Oddly
enough you can at least get a system booted on 17h family systems even
if it's fragile but 15h just fails to even boot.  Not sure how it even
worked when I tested previous patches on the 15h system. 

-- 
Brian Woods

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