Re: [Qemu-devel] i386: pc: align gpa<->hpa on 1GB boundary (v4)

From: "Michael S. Tsirkin" <mst@redhat.com>
To: Marcelo Tosatti <mtosatti@redhat.com>
Cc: aarcange@redhat.com, gleb@redhat.com, qemu-devel@nongnu.org,
	Gerd Hoffmann <kraxel@redhat.com>,
	Igor Mammedov <imammedo@redhat.com>,
	pbonzini@redhat.com
Subject: Re: [Qemu-devel] i386: pc: align gpa<->hpa on 1GB boundary (v4)
Date: Thu, 7 Nov 2013 00:15:59 +0200	[thread overview]
Message-ID: <20131106221559.GA22730@redhat.com> (raw)
In-Reply-To: <20131106215351.GB15707@amt.cnet>

On Wed, Nov 06, 2013 at 07:53:51PM -0200, Marcelo Tosatti wrote:
> On Wed, Nov 06, 2013 at 11:40:34PM +0200, Michael S. Tsirkin wrote:
> > On Wed, Nov 06, 2013 at 07:31:19PM -0200, Marcelo Tosatti wrote:
> > > 
> > > v2: condition enablement of new mapping to new machine types (Paolo)
> > > v3: fix changelog
> > > v4: rebase
> > > 
> > > -----
> > > 
> > > 
> > > Align guest physical address and host physical address
> > > beyond guest 4GB on a 1GB boundary.
> > > 
> > > Otherwise 1GB TLBs cannot be cached for the range.
> > > 
> > > Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
> > 
> > Um. This will conflict with:
> >     pc: map PCI address space as catchall region for not mapped addresses
> > 
> > I think we really should stop using the hacked hole thing
> > and just use priorities like that patch does.
> 
> Sorry hacked in what way?
> This patch is necessary to enable 1GB hugepages beyond 4GB of RAM on the
> current machine types.

Sorry if I wasn't clear. when I said "hacked" I was talking about the
pci hole concept generally in upstream qemu, not about your patch.

Its hacked because there's no "pci hole" on PIIX.
pci hole is where pci was hiding some ram behind it
on some systems. AFAIK this is not what is happens on piix though.
What happens really is that everything not covered by RAM memory is PCI.

We implemented this using two aliases of RAM but
the natural thing is really just making PCI lower
priority than RAM and let it overlap.

> > Do you agree? If yes I'm afraid your patch will have to be
> > rebased on top of that yet again, sorry to give you a
> > run-around like that :(
> 
> I don't see what exactly is the suggestion (or why the proposed 
> patch should conflict with "pc: map PCI address space as catchall region
> for not mapped addresses").

It seemed to me that they will conflict but it's after midnight
so maybe I'm confused.
You are saying you apply yours on top and there's no conflict?
In that case I'll recheck, sorry.

> > Also - do you think this is 1.7 material?
> 
> No. Paolo mentioned you have a tree with 1.8 material, correct?

Yes

git://git.kernel.org/pub/scm/virt/kvm/mst/qemu.git pci