From: Yu Zhang <yu.c.zhang@linux.intel.com>
To: Andy Lutomirski <luto@kernel.org>
Cc: David Hildenbrand <david@redhat.com>,
Sean Christopherson <seanjc@google.com>,
Paolo Bonzini <pbonzini@redhat.com>,
Vitaly Kuznetsov <vkuznets@redhat.com>,
Wanpeng Li <wanpengli@tencent.com>,
Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
kvm list <kvm@vger.kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
Borislav Petkov <bp@alien8.de>,
Andrew Morton <akpm@linux-foundation.org>,
Joerg Roedel <jroedel@suse.de>, Andi Kleen <ak@linux.intel.com>,
David Rientjes <rientjes@google.com>,
Vlastimil Babka <vbabka@suse.cz>,
Tom Lendacky <thomas.lendacky@amd.com>,
Thomas Gleixner <tglx@linutronix.de>,
"Peter Zijlstra (Intel)" <peterz@infradead.org>,
Ingo Molnar <mingo@redhat.com>,
Varad Gautam <varad.gautam@suse.com>,
Dario Faggioli <dfaggioli@suse.com>,
the arch/x86 maintainers <x86@kernel.org>,
linux-mm@kvack.org, linux-coco@lists.linux.dev,
"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
"Kirill A . Shutemov" <kirill@shutemov.name>,
Sathyanarayanan Kuppuswamy
<sathyanarayanan.kuppuswamy@linux.intel.com>,
Dave Hansen <dave.hansen@intel.com>
Subject: Re: [RFC] KVM: mm: fd-based approach for supporting KVM guest private memory
Date: Thu, 2 Sep 2021 16:19:23 +0800 [thread overview]
Message-ID: <20210902081923.lertnjsgnskegkmn@linux.intel.com> (raw)
In-Reply-To: <f37a61ba-b7ef-c789-5763-f7f237ae41cc@kernel.org>
On Wed, Sep 01, 2021 at 09:07:59AM -0700, Andy Lutomirski wrote:
> On 9/1/21 3:24 AM, Yu Zhang wrote:
> > On Tue, Aug 31, 2021 at 09:53:27PM -0700, Andy Lutomirski wrote:
> >>
> >>
> >> On Thu, Aug 26, 2021, at 7:31 PM, Yu Zhang wrote:
> >>> On Thu, Aug 26, 2021 at 12:15:48PM +0200, David Hildenbrand wrote:
> >>
> >>> Thanks a lot for this summary. A question about the requirement: do we or
> >>> do we not have plan to support assigned device to the protected VM?
> >>>
> >>> If yes. The fd based solution may need change the VFIO interface as well(
> >>> though the fake swap entry solution need mess with VFIO too). Because:
> >>>
> >>> 1> KVM uses VFIO when assigning devices into a VM.
> >>>
> >>> 2> Not knowing which GPA ranges may be used by the VM as DMA buffer, all
> >>> guest pages will have to be mapped in host IOMMU page table to host pages,
> >>> which are pinned during the whole life cycle fo the VM.
> >>>
> >>> 3> IOMMU mapping is done during VM creation time by VFIO and IOMMU driver,
> >>> in vfio_dma_do_map().
> >>>
> >>> 4> However, vfio_dma_do_map() needs the HVA to perform a GUP to get the HPA
> >>> and pin the page.
> >>>
> >>> But if we are using fd based solution, not every GPA can have a HVA, thus
> >>> the current VFIO interface to map and pin the GPA(IOVA) wont work. And I
> >>> doubt if VFIO can be modified to support this easily.
> >>>
> >>>
> >>
> >> Do you mean assigning a normal device to a protected VM or a hypothetical protected-MMIO device?
> >>
> >> If the former, it should work more or less like with a non-protected VM. mmap the VFIO device, set up a memslot, and use it. I'm not sure whether anyone will actually do this, but it should be possible, at least in principle. Maybe someone will want to assign a NIC to a TDX guest. An NVMe device with the understanding that the guest can't trust it wouldn't be entirely crazy ether.
> >>
> >> If the latter, AFAIK there is no spec for how it would work even in principle. Presumably it wouldn't work quite like VFIO -- instead, the kernel could have a protection-virtual-io-fd mechanism, and that fd could be bound to a memslot in whatever way we settle on for binding secure memory to a memslot.
> >>
> >
> > Thanks Andy. I was asking the first scenario.
> >
> > Well, I agree it is doable if someone really want some assigned
> > device in TD guest. As Kevin mentioned in his reply, HPA can be
> > generated, by extending VFIO with a new mapping protocol which
> > uses fd+offset, instead of HVA.
>
> I'm confused. I don't see why any new code is needed for this at all.
> Every proposal I've seen for handling TDX memory continues to handle TDX
> *shared* memory exactly like regular guest memory today. The only
> differences are that more hole punching will be needed, which will
> require lightweight memslots (to have many of them), memslots with
> holes, or mappings backing memslots with holes (which can be done with
> munmap() on current kernels).
Thanks for pointing this out. And yes, for DMAs not capable of encryption(
which is the case in current TDX). GUP shall work as it is in VFIO. :)
>
> So you can literally just mmap a VFIO device and expect it to work,
> exactly like it does right now. Whether the guest will be willing to
> use the device will depend on the guest security policy (all kinds of
> patches about that are flying around), but if the guest tries to use it,
> it really should just work.
>
But I think there's still problem. For now,
1> Qemu mmap()s all GPAs into its HVA space, when the VM is created.
2> With no idea which part of guest memory shall be shared, VFIO will just
set up the IOPT, by mapping whole GPA ranges in IOPT.
3> And those GPAs are actually private ones, with no shared-bit set.
Later when guest tries to perform a DMA(using a shared GPA), IO page fault
shall happen.
> >
> > Another issue is current TDX does not support DMA encryption, and
> > only shared GPA memory shall be mapped in the VT-d. So to support
> > this, KVM may need to work with VFIO to dynamically program host
> > IOPT(IOMMU Page Table) when TD guest notifies a shared GFN range(e.g.,
> > with a MAP_GPA TDVMCALL), instead of prepopulating the IOPT at VM
> > creation time, by mapping entire GFN ranges of a guest.
>
> Given that there is no encrypted DMA support, shouldn't the only IOMMU
> mappings (real host-side IOMMU) that point at guest memory be for
> non-encrypted DMA? I don't see how this interacts at all. If the guest
> tries to MapGPA to turn a shared MMIO page into private, the host should
> fail the hypercall because the operation makes no sense.
>
> It is indeed the case that, with a TDX guest, MapGPA shared->private to
> a page that was previously used for unencrypted DMA will need to avoid
> having IOPT entries to the new private page, but even that doesn't seem
> particularly bad. The fd+special memslot proposal for private memory
> means that shared *backing store* pages never actually transition
> between shared and private without being completely freed.
>
> As far as I can tell, the actual problem you're referring to is:
>
> >>> 2> Not knowing which GPA ranges may be used by the VM as DMA buffer, all
> >>> guest pages will have to be mapped in host IOMMU page table to host
> pages,
> >>> which are pinned during the whole life cycle fo the VM.
Yes. That's the primary concern. :)
>
> In principle, you could actually initialize a TDX guest with all of its
> memory shared and all of it mapped in the host IOMMU. When a guest
> turns some pages private, user code could punch a hole in the memslot,
> allocate private memory at that address, but leave the shared backing
> store in place and still mapped in the host IOMMU. The result would be
> that guest-initiated DMA to the previously shared address would actually
> work but would hit pages that are invisible to the guest. And a whole
> bunch of memory would be waste, but the whole system should stll work.
Do you mean to let VFIO & IOMMU to treat all guest memory as shared first,
and then just allocate the private pages in another backing store? I guess
that could work, but with the cost of allocating roughly 2x physical pages
of the guest RAM size. After all, the shared pages shall be only a small
part of guest memory.
If device assignment is desired in current TDX. My understanding of the
enabling work would be like this:
1> Change qemu to not trigger VFIO_IOMMU_MAP_DMA for the TD, thus no IOPT
prepopulated, and no physical page allocated.
2> KVM forwards MapGPA(private -> shared) request to Qemu.
3> Qemu asks VFIO to pin and map the shared GPAs.
For private -> shared transitions, the memslot punching, IOPT unmapping,
and iotlb flushing are necessary. Possibly new interface between VFIO and
KVM is needed.
But actually I am not sure if people really want assigned device in current
TDX. Bottleneck of the performance should be the copying to/from swiotlb
buffers.
B.R.
Yu
next prev parent reply other threads:[~2021-09-02 8:19 UTC|newest]
Thread overview: 71+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-24 0:52 [RFC] KVM: mm: fd-based approach for supporting KVM guest private memory Sean Christopherson
2021-08-24 10:48 ` Yu Zhang
2021-08-26 0:35 ` Sean Christopherson
2021-08-26 13:23 ` Yu Zhang
2021-08-26 10:15 ` David Hildenbrand
2021-08-26 17:05 ` Andy Lutomirski
2021-08-26 21:26 ` David Hildenbrand
2021-08-27 18:24 ` Andy Lutomirski
2021-08-27 22:28 ` Sean Christopherson
2021-08-31 19:12 ` David Hildenbrand
2021-08-31 20:45 ` Sean Christopherson
2021-09-01 7:51 ` David Hildenbrand
2021-08-27 2:31 ` Yu Zhang
2021-08-31 19:08 ` David Hildenbrand
2021-08-31 20:01 ` Andi Kleen
2021-08-31 20:15 ` David Hildenbrand
2021-08-31 20:39 ` Andi Kleen
2021-09-01 3:34 ` Yu Zhang
2021-09-01 4:53 ` Andy Lutomirski
2021-09-01 7:12 ` Tian, Kevin
2021-09-01 10:24 ` Yu Zhang
2021-09-01 16:07 ` Andy Lutomirski
2021-09-01 16:27 ` David Hildenbrand
2021-09-02 8:34 ` Yu Zhang
2021-09-02 8:44 ` David Hildenbrand
2021-09-02 11:02 ` Yu Zhang
2021-09-02 8:19 ` Yu Zhang [this message]
2021-09-02 18:41 ` Andy Lutomirski
2021-09-07 1:33 ` Yan Zhao
2021-09-02 9:27 ` Joerg Roedel
2021-09-02 18:41 ` Andy Lutomirski
2021-09-02 18:57 ` Sean Christopherson
2021-09-02 19:07 ` Dave Hansen
2021-09-02 20:42 ` Andy Lutomirski
2021-08-27 22:18 ` Sean Christopherson
2021-08-31 19:07 ` David Hildenbrand
2021-08-31 21:54 ` Sean Christopherson
2021-09-01 8:09 ` David Hildenbrand
2021-09-01 15:54 ` Andy Lutomirski
2021-09-01 16:16 ` David Hildenbrand
2021-09-01 17:09 ` Andy Lutomirski
2021-09-01 16:18 ` James Bottomley
2021-09-01 16:22 ` David Hildenbrand
2021-09-01 16:31 ` James Bottomley
2021-09-01 16:37 ` David Hildenbrand
2021-09-01 16:45 ` James Bottomley
2021-09-01 17:08 ` David Hildenbrand
2021-09-01 17:50 ` Sean Christopherson
2021-09-01 17:53 ` David Hildenbrand
2021-09-01 17:08 ` Andy Lutomirski
2021-09-01 17:13 ` James Bottomley
2021-09-02 10:18 ` Joerg Roedel
2021-09-01 18:24 ` Andy Lutomirski
2021-09-01 19:26 ` Dave Hansen
2021-09-07 15:00 ` Tom Lendacky
2021-09-01 4:58 ` Andy Lutomirski
2021-09-01 7:49 ` David Hildenbrand
2021-09-02 18:47 ` Kirill A. Shutemov
2021-09-02 20:33 ` Sean Christopherson
2021-09-03 19:14 ` Kirill A. Shutemov
2021-09-03 19:15 ` Andy Lutomirski
2021-09-10 17:18 ` Kirill A. Shutemov
2021-09-15 19:58 ` Chao Peng
2021-09-15 13:51 ` David Hildenbrand
2021-09-15 14:29 ` Kirill A. Shutemov
2021-09-15 14:59 ` David Hildenbrand
2021-09-15 15:35 ` David Hildenbrand
2021-09-15 20:04 ` Kirill A. Shutemov
2021-09-15 14:11 ` Kirill A. Shutemov
2021-09-16 7:36 ` Chao Peng
2021-09-16 9:24 ` Paolo Bonzini
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210902081923.lertnjsgnskegkmn@linux.intel.com \
--to=yu.c.zhang@linux.intel.com \
--cc=ak@linux.intel.com \
--cc=akpm@linux-foundation.org \
--cc=bp@alien8.de \
--cc=dave.hansen@intel.com \
--cc=david@redhat.com \
--cc=dfaggioli@suse.com \
--cc=jmattson@google.com \
--cc=joro@8bytes.org \
--cc=jroedel@suse.de \
--cc=kirill.shutemov@linux.intel.com \
--cc=kirill@shutemov.name \
--cc=kvm@vger.kernel.org \
--cc=linux-coco@lists.linux.dev \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=luto@kernel.org \
--cc=mingo@redhat.com \
--cc=pbonzini@redhat.com \
--cc=peterz@infradead.org \
--cc=rientjes@google.com \
--cc=sathyanarayanan.kuppuswamy@linux.intel.com \
--cc=seanjc@google.com \
--cc=tglx@linutronix.de \
--cc=thomas.lendacky@amd.com \
--cc=varad.gautam@suse.com \
--cc=vbabka@suse.cz \
--cc=vkuznets@redhat.com \
--cc=wanpengli@tencent.com \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).