Re: [PATCH v2] x86/mm: Add mem access rights to NPT

From: Alexandru Stefan ISAILA <aisaila@bitdefender.com>
To: "JBeulich@suse.com" <JBeulich@suse.com>,
	"tamas@tklengyel.com" <tamas@tklengyel.com>
Cc: "George.Dunlap@eu.citrix.com" <George.Dunlap@eu.citrix.com>,
	"andrew.cooper3@citrix.com" <andrew.cooper3@citrix.com>,
	"rcojocaru@bitdefender.com" <rcojocaru@bitdefender.com>,
	"xen-devel@lists.xen.org" <xen-devel@lists.xen.org>
Subject: Re: [PATCH v2] x86/mm: Add mem access rights to NPT
Date: Mon, 2 Jul 2018 08:01:36 +0000	[thread overview]
Message-ID: <1530518521.8880.21.camel@bitdefender.com> (raw)
In-Reply-To: <5B39C83302000078001CF8E2@prv1-mh.provo.novell.com>

On Lu, 2018-07-02 at 00:37 -0600, Jan Beulich wrote:
> >
> > >
> > > >
> > > > On 29.06.18 at 18:42, <tamas@tklengyel.com> wrote:
> > On Fri, Jun 29, 2018 at 3:38 AM Jan Beulich <JBeulich@suse.com>
> > wrote:
> > >
> > >
> > > >
> > > > >
> > > > > >
> > > > > > On 29.06.18 at 10:42, <aisaila@bitdefender.com> wrote:
> > > > I've started a win 7 machine with no introspection or xen-
> > > > access. After
> > > > 30 min of uptime there were 0 inserts into the tree. I guess
> > > > the
> > > > overhead is down to a minimum with no user modified access
> > > > rights.
> > > Thanks, this is helpful to know. In that case though, as said
> > > before, I'd
> > > like to ask to defer setting up of the radix tree to the point
> > > where it
> > > actually is going to be needed.
> > Are you worried about the radix tree being present (ie
> > radix_tree_init
> > have been called) before it is used for mem_access? As I said
> > earlier,
> > a lookup on an empty radix tree is equivalent of a NULL check. I
> > don't
> > get how not having the tree initialized will be any faster then
> > doing
> > a lookup on an empty one.
> No, the question is not about performance. The point is about
> reassurance that the tree isn't going to be used in normal (non-
> introspection) operation. If it suddenly and unknowingly became
> used down the road, the resource consumption pattern of
> domains may change quite significantly. Not setting up the tree
> unless needed likely also helps review of the changes, as it'll be
> necessary to make sure in the patch that it won't get accessed
> without having been set up.
>
I can move the radix tree init to p2m_pt_set_entry() so that it is
called on the first use or I can move it to vm_event_enable() and then
return in every case that the root is NULL.

What would be the best way to handle this?

Thanks,
Alex

________________________
This email was scanned by Bitdefender
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel