Re: [PATCH v1] drm/ttm: Refcount allocated tail pages

From: Sean Christopherson <seanjc@google.com>
To: "Christian König" <christian.koenig@amd.com>,
	"Dmitry Osipenko" <dmitry.osipenko@collabora.com>,
	"David Airlie" <airlied@linux.ie>,
	"Huang Rui" <ray.huang@amd.com>,
	"Trigger Huang" <Trigger.Huang@gmail.com>,
	"Gert Wollny" <gert.wollny@collabora.com>,
	"Antonio Caggiano" <antonio.caggiano@collabora.com>,
	dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org,
	"Dmitry Osipenko" <digetx@gmail.com>,
	kvm@vger.kernel.org, kernel@collabora.com,
	virtualization@lists.linux-foundation.org
Subject: Re: [PATCH v1] drm/ttm: Refcount allocated tail pages
Date: Wed, 11 Jan 2023 17:13:23 +0000	[thread overview]
Message-ID: <Y77uM/X94DtKXtK0@google.com> (raw)
In-Reply-To: <YxeoEr6xAtlZ+IrU@phenom.ffwll.local>

On Tue, Sep 06, 2022, Daniel Vetter wrote:
> On Tue, Sep 06, 2022 at 10:01:47PM +0200, Daniel Vetter wrote:
> > On Mon, Aug 15, 2022 at 12:05:19PM +0200, Christian König wrote:
> > > Am 15.08.22 um 11:54 schrieb Dmitry Osipenko:
> > > > Higher order pages allocated using alloc_pages() aren't refcounted and they
> > > > need to be refcounted, otherwise it's impossible to map them by KVM. This
> > > > patch sets the refcount of the tail pages and fixes the KVM memory mapping
> > > > faults.
> > > > 
> > > > Without this change guest virgl driver can't map host buffers into guest
> > > > and can't provide OpenGL 4.5 profile support to the guest. The host
> > > > mappings are also needed for enabling the Venus driver using host GPU
> > > > drivers that are utilizing TTM.
> > > > 
> > > > Based on a patch proposed by Trigger Huang.
> > > 
> > > Well I can't count how often I have repeated this: This is an absolutely
> > > clear NAK!
> > > 
> > > TTM pages are not reference counted in the first place and because of this
> > > giving them to virgl is illegal.
> > > 
> > > Please immediately stop this completely broken approach. We have discussed
> > > this multiple times now.
> > 
> > Yeah we need to get this stuff closed for real by tagging them all with
> > VM_IO or VM_PFNMAP asap.
> 
> For a bit more context: Anything mapping a bo should be VM_SPECIAL. And I
> think we should add the checks to the gem and dma-buf mmap functions to
> validate for that, and fix all the fallout.
> 
> Otherwise this dragon keeps resurrecting ...
> 
> VM_SPECIAL _will_ block get_user_pages, which will block everyone from
> even trying to refcount this stuff.

FWIW, IIUC that won't change the KVM story.  KVM acquires the PFN for these pages
via follow_pte(), not by gup().  Details are in a different strand of this thread[*].

If TTM pages aren't tied into mmu_notifiers, then I believe the only solution is
to not allow them to be mapped into user page tables.  If they are tied into
mmu_notifiers, then this is fully a KVM limitation that we are (slowly) resolving.

[*] https://lore.kernel.org/all/Y77sQZI0IfFVx7Jo@google.com