Message-ID: <20220530132613.GA1200843@chaop.bj.intel.com>
Date:   Mon, 30 May 2022 21:26:13 +0800
From:   Chao Peng <chao.p.peng@...ux.intel.com>
To:     Sean Christopherson <seanjc@...gle.com>
Cc:     Andy Lutomirski <luto@...nel.org>, kvm@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        linux-fsdevel@...r.kernel.org, linux-api@...r.kernel.org,
        linux-doc@...r.kernel.org, qemu-devel@...gnu.org,
        Paolo Bonzini <pbonzini@...hat.com>,
        Jonathan Corbet <corbet@....net>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Jim Mattson <jmattson@...gle.com>,
        Joerg Roedel <joro@...tes.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
        x86@...nel.org, "H . Peter Anvin" <hpa@...or.com>,
        Hugh Dickins <hughd@...gle.com>,
        Jeff Layton <jlayton@...nel.org>,
        "J . Bruce Fields" <bfields@...ldses.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Mike Rapoport <rppt@...nel.org>,
        Steven Price <steven.price@....com>,
        "Maciej S . Szmigiero" <mail@...iej.szmigiero.name>,
        Vlastimil Babka <vbabka@...e.cz>,
        Vishal Annapurve <vannapurve@...gle.com>,
        Yu Zhang <yu.c.zhang@...ux.intel.com>,
        "Kirill A . Shutemov" <kirill.shutemov@...ux.intel.com>,
        jun.nakajima@...el.com, dave.hansen@...el.com, ak@...ux.intel.com,
        david@...hat.com, aarcange@...hat.com, ddutile@...hat.com,
        dhildenb@...hat.com, Quentin Perret <qperret@...gle.com>,
        Michael Roth <michael.roth@....com>, mhocko@...e.com
Subject: Re: [PATCH v6 4/8] KVM: Extend the memslot to support fd-based
 private memory

On Mon, May 23, 2022 at 03:22:32PM +0000, Sean Christopherson wrote:
> On Mon, May 23, 2022, Chao Peng wrote:
> > On Fri, May 20, 2022 at 06:31:02PM +0000, Sean Christopherson wrote:
> > > On Fri, May 20, 2022, Andy Lutomirski wrote:
> > > > The alternative would be to have some kind of separate table or bitmap (part
> > > > of the memslot?) that tells KVM whether a GPA should map to the fd.
> > > > 
> > > > What do you all think?
> > > 
> > > My original proposal was to have expolicit shared vs. private memslots, and punch
> > > holes in KVM's memslots on conversion, but due to the way KVM (and userspace)
> > > handle memslot updates, conversions would be painfully slow.  That's how we ended
> > > up with the current propsoal.
> > > 
> > > But a dedicated KVM ioctl() to add/remove shared ranges would be easy to implement
> > > and wouldn't necessarily even need to interact with the memslots.  It could be a
> > > consumer of memslots, e.g. if we wanted to disallow registering regions without an
> > > associated memslot, but I think we'd want to avoid even that because things will
> > > get messy during memslot updates, e.g. if dirty logging is toggled or a shared
> > > memory region is temporarily removed then we wouldn't want to destroy the tracking.
> > 
> > Even we don't tight that to memslots, that info can only be effective
> > for private memslot, right? Setting this ioctl to memory ranges defined
> > in a traditional non-private memslots just makes no sense, I guess we can
> > comment that in the API document.
> 
> Hrm, applying it universally would be funky, e.g. emulated MMIO would need to be
> declared "shared".  But, applying it selectively would arguably be worse, e.g.
> letting userspace map memory into the guest as shared for a region that's registered
> as private...
> 
> On option to that mess would be to make memory shared by default, and so userspace
> must declare regions that are private.  Then there's no weirdness with emulated MMIO
> or "legacy" memslots.
> 
> On page fault, KVM does a lookup to see if the GPA is shared or private.  If the
> GPA is private, but there is no memslot or the memslot doesn't have a private fd,
> KVM exits to userspace.  If there's a memslot with a private fd, the shared/private
> flag is used to resolve the 
> 
> And to handle the ioctl(), KVM can use kvm_zap_gfn_range(), which will bump the
> notifier sequence, i.e. force the page fault to retry if the GPA may have been
> (un)registered between checking the type and acquiring mmu_lock.

Yeah, that makes sense.

> 
> > > I don't think we'd want to use a bitmap, e.g. for a well-behaved guest, XArray
> > > should be far more efficient.
> > 
> > What about the mis-behaved guest? I don't want to design for the worst
> > case, but people may raise concern on the attack from such guest.
> 
> That's why cgroups exist.  E.g. a malicious/broken L1 can similarly abuse nested
> EPT/NPT to generate a large number of shadow page tables.

I havn't seen we had that in KVM. Is there any plan/discussion to add that?

> 
> > > One benefit to explicitly tracking this in KVM is that it might be useful for
> > > software-only protected VMs, e.g. KVM could mark a region in the XArray as "pending"
> > > based on guest hypercalls to share/unshare memory, and then complete the transaction
> > > when userspace invokes the ioctl() to complete the share/unshare.
> > 
> > OK, then this can be another field of states/flags/attributes. Let me
> > dig up certain level of details:
> > 
> > First, introduce below KVM ioctl
> > 
> > KVM_SET_MEMORY_ATTR
> 
> Actually, if the semantics are that userspace declares memory as private, then we
> can reuse KVM_MEMORY_ENCRYPT_REG_REGION and KVM_MEMORY_ENCRYPT_UNREG_REGION.  It'd
> be a little gross because we'd need to slightly redefine the semantics for TDX, SNP,
> and software-protected VM types, e.g. the ioctls() currently require a pre-exisitng
> memslot.  But I think it'd work...

These existing ioctls looks good for TDX and probably SNP as well. For
softrware-protected VM types, it may not be enough. Maybe for the first
step we can reuse this for all hardware based solutions and invent new
interface when software-protected solution gets really supported.

There is semantics difference for fd-based private memory. Current above
two ioctls() use userspace addreess(hva) while for fd-based it should be
fd+offset, and probably it's better to use gpa in this case. Then we
will need change existing semantics and break backward-compatibility.

Chao

> 
> I'll think more on this...

Powered by blists - more mailing lists