Subject: Re: [PATCH Part2 v5 00/45] Add AMD Secure Nested Paging (SEV-SNP)
 Hypervisor Support
To:     Peter Gonda <pgonda@google.com>,
        Brijesh Singh <brijesh.singh@amd.com>
Cc:     x86@kernel.org, linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        linux-coco@lists.linux.dev, linux-mm@kvack.org,
        linux-crypto@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Joerg Roedel <jroedel@suse.de>,
        Tom Lendacky <Thomas.Lendacky@amd.com>,
        "H. Peter Anvin" <hpa@zytor.com>, Ard Biesheuvel <ardb@kernel.org>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Sean Christopherson <seanjc@google.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Andy Lutomirski <luto@kernel.org>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Sergio Lopez <slp@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>,
        David Rientjes <rientjes@google.com>,
        Dov Murik <dovmurik@linux.ibm.com>,
        Tobin Feldman-Fitzthum <tobin@ibm.com>,
        Borislav Petkov <bp@alien8.de>,
        Michael Roth <michael.roth@amd.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        "Kirill A . Shutemov" <kirill@shutemov.name>,
        Andi Kleen <ak@linux.intel.com>, tony.luck@intel.com,
        marcorr@google.com, sathyanarayanan.kuppuswamy@linux.intel.com
References: <20210820155918.7518-1-brijesh.singh@amd.com>
 <CAMkAt6o0ySn1=iLYsH0LCnNARrUbfaS0cvtxB__y_d+Q6DUzfA@mail.gmail.com>
From:   Dave Hansen <dave.hansen@intel.com>
Autocrypt: addr=dave.hansen@intel.com; keydata=
 xsFNBE6HMP0BEADIMA3XYkQfF3dwHlj58Yjsc4E5y5G67cfbt8dvaUq2fx1lR0K9h1bOI6fC
 oAiUXvGAOxPDsB/P6UEOISPpLl5IuYsSwAeZGkdQ5g6m1xq7AlDJQZddhr/1DC/nMVa/2BoY
 2UnKuZuSBu7lgOE193+7Uks3416N2hTkyKUSNkduyoZ9F5twiBhxPJwPtn/wnch6n5RsoXsb
 ygOEDxLEsSk/7eyFycjE+btUtAWZtx+HseyaGfqkZK0Z9bT1lsaHecmB203xShwCPT49Blxz
 VOab8668QpaEOdLGhtvrVYVK7x4skyT3nGWcgDCl5/Vp3TWA4K+IofwvXzX2ON/Mj7aQwf5W
 iC+3nWC7q0uxKwwsddJ0Nu+dpA/UORQWa1NiAftEoSpk5+nUUi0WE+5DRm0H+TXKBWMGNCFn
 c6+EKg5zQaa8KqymHcOrSXNPmzJuXvDQ8uj2J8XuzCZfK4uy1+YdIr0yyEMI7mdh4KX50LO1
 pmowEqDh7dLShTOif/7UtQYrzYq9cPnjU2ZW4qd5Qz2joSGTG9eCXLz5PRe5SqHxv6ljk8mb
 ApNuY7bOXO/A7T2j5RwXIlcmssqIjBcxsRRoIbpCwWWGjkYjzYCjgsNFL6rt4OL11OUF37wL
 QcTl7fbCGv53KfKPdYD5hcbguLKi/aCccJK18ZwNjFhqr4MliQARAQABzShEYXZpZCBDaHJp
 c3RvcGhlciBIYW5zZW4gPGRhdmVAc3I3MS5uZXQ+wsF7BBMBAgAlAhsDBgsJCAcDAgYVCAIJ
 CgsEFgIDAQIeAQIXgAUCTo3k0QIZAQAKCRBoNZUwcMmSsMO2D/421Xg8pimb9mPzM5N7khT0
 2MCnaGssU1T59YPE25kYdx2HntwdO0JA27Wn9xx5zYijOe6B21ufrvsyv42auCO85+oFJWfE
 K2R/IpLle09GDx5tcEmMAHX6KSxpHmGuJmUPibHVbfep2aCh9lKaDqQR07gXXWK5/yU1Dx0r
 VVFRaHTasp9fZ9AmY4K9/BSA3VkQ8v3OrxNty3OdsrmTTzO91YszpdbjjEFZK53zXy6tUD2d
 e1i0kBBS6NLAAsqEtneplz88T/v7MpLmpY30N9gQU3QyRC50jJ7LU9RazMjUQY1WohVsR56d
 ORqFxS8ChhyJs7BI34vQusYHDTp6PnZHUppb9WIzjeWlC7Jc8lSBDlEWodmqQQgp5+6AfhTD
 kDv1a+W5+ncq+Uo63WHRiCPuyt4di4/0zo28RVcjtzlGBZtmz2EIC3vUfmoZbO/Gn6EKbYAn
 rzz3iU/JWV8DwQ+sZSGu0HmvYMt6t5SmqWQo/hyHtA7uF5Wxtu1lCgolSQw4t49ZuOyOnQi5
 f8R3nE7lpVCSF1TT+h8kMvFPv3VG7KunyjHr3sEptYxQs4VRxqeirSuyBv1TyxT+LdTm6j4a
 mulOWf+YtFRAgIYyyN5YOepDEBv4LUM8Tz98lZiNMlFyRMNrsLV6Pv6SxhrMxbT6TNVS5D+6
 UorTLotDZKp5+M7BTQRUY85qARAAsgMW71BIXRgxjYNCYQ3Xs8k3TfAvQRbHccky50h99TUY
 sqdULbsb3KhmY29raw1bgmyM0a4DGS1YKN7qazCDsdQlxIJp9t2YYdBKXVRzPCCsfWe1dK/q
 66UVhRPP8EGZ4CmFYuPTxqGY+dGRInxCeap/xzbKdvmPm01Iw3YFjAE4PQ4hTMr/H76KoDbD
 cq62U50oKC83ca/PRRh2QqEqACvIH4BR7jueAZSPEDnzwxvVgzyeuhwqHY05QRK/wsKuhq7s
 UuYtmN92Fasbxbw2tbVLZfoidklikvZAmotg0dwcFTjSRGEg0Gr3p/xBzJWNavFZZ95Rj7Et
 db0lCt0HDSY5q4GMR+SrFbH+jzUY/ZqfGdZCBqo0cdPPp58krVgtIGR+ja2Mkva6ah94/oQN
 lnCOw3udS+Eb/aRcM6detZr7XOngvxsWolBrhwTQFT9D2NH6ryAuvKd6yyAFt3/e7r+HHtkU
 kOy27D7IpjngqP+b4EumELI/NxPgIqT69PQmo9IZaI/oRaKorYnDaZrMXViqDrFdD37XELwQ
 gmLoSm2VfbOYY7fap/AhPOgOYOSqg3/Nxcapv71yoBzRRxOc4FxmZ65mn+q3rEM27yRztBW9
 AnCKIc66T2i92HqXCw6AgoBJRjBkI3QnEkPgohQkZdAb8o9WGVKpfmZKbYBo4pEAEQEAAcLB
 XwQYAQIACQUCVGPOagIbDAAKCRBoNZUwcMmSsJeCEACCh7P/aaOLKWQxcnw47p4phIVR6pVL
 e4IEdR7Jf7ZL00s3vKSNT+nRqdl1ugJx9Ymsp8kXKMk9GSfmZpuMQB9c6io1qZc6nW/3TtvK
 pNGz7KPPtaDzvKA4S5tfrWPnDr7n15AU5vsIZvgMjU42gkbemkjJwP0B1RkifIK60yQqAAlT
 YZ14P0dIPdIPIlfEPiAWcg5BtLQU4Wg3cNQdpWrCJ1E3m/RIlXy/2Y3YOVVohfSy+4kvvYU3
 lXUdPb04UPw4VWwjcVZPg7cgR7Izion61bGHqVqURgSALt2yvHl7cr68NYoFkzbNsGsye9ft
 M9ozM23JSgMkRylPSXTeh5JIK9pz2+etco3AfLCKtaRVysjvpysukmWMTrx8QnI5Nn5MOlJj
 1Ov4/50JY9pXzgIDVSrgy6LYSMc4vKZ3QfCY7ipLRORyalFDF3j5AGCMRENJjHPD6O7bl3Xo
 4DzMID+8eucbXxKiNEbs21IqBZbbKdY1GkcEGTE7AnkA3Y6YB7I/j9mQ3hCgm5muJuhM/2Fr
 OPsw5tV/LmQ5GXH0JQ/TZXWygyRFyyI2FqNTx4WHqUn3yFj8rwTAU1tluRUYyeLy0ayUlKBH
 ybj0N71vWO936MqP6haFERzuPAIpxj2ezwu0xb1GjTk4ynna6h5GjnKgdfOWoRtoWndMZxbA
 z5cecg==
Message-ID: <061ccd49-3b9f-d603-bafd-61a067c3f6fa@intel.com>
Date:   Fri, 12 Nov 2021 09:59:46 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <CAMkAt6o0ySn1=iLYsH0LCnNARrUbfaS0cvtxB__y_d+Q6DUzfA@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 11/12/21 7:43 AM, Peter Gonda wrote:
...
> Here is an alternative to the current approach: On RMP violation (host
> or userspace) the page fault handler converts the page from private to
> shared to allow the write to continue. This pulls from s390’s error
> handling which does exactly this. See ‘arch_make_page_accessible()’.
> Additionally it adds less complexity to the SNP kernel patches, and
> requires no new ABI.

I think it's important to very carefully describe where these RMP page
faults can occur within the kernel.  They can obvious occur on things like:

	copy_to_user(&user_buf, &kernel_buf, len);

That's not a big deal.  Those can obviously handle page faults.  We know
exactly the instruction on which the fault can occur and we handle it
gracefully.

*But*, these are harder:

	get_user_pages(addr, len, &pages);
	spin_lock(&lock);
	ptr = kmap_atomic(pages[0]);
	*ptr = foo; // #PF here
	kunmap_atomic(ptr);
	spin_unlock(&lock);
	put_page(pages[0]);

In this case, the place where the fault happens are not especially well
bounded.  It can be in compiler-generated code.  It can happen on any
random instruction in there.

Or, is there some mechanism that prevent guest-private memory from being
accessed in random host kernel code?

> This proposal does cause guest memory corruption for some bugs but one
> of SEV-SNP’s goals extended from SEV-ES’s goals is for guest’s to be
> able to detect when its memory has been corrupted / replayed by the
> host. So SNP already has features for allowing guests to detect this
> kind of memory corruption. Additionally this is very similar to a page
> of memory generating a machine check because of 2-bit memory
> corruption. In other words SNP guests must be enlightened and ready
> for these kinds of errors.
> 
> For an SNP guest running under this proposal the flow would look like this:
> * Host gets a #PF because its trying to write to a private page.
> * Host #PF handler updates the page to shared.
> * Write continues normally.
> * Guest accesses memory (r/w).
> * Guest gets a #VC error because the page is not PVALIDATED
> * Guest is now in control. Guest can terminate because its memory has
> been corrupted. Guest could try and continue to log the error to its
> owner.

This sounds like a _possible_ opportunity for the guest to do some extra
handling.  It's also quite possible that this #VC happens in a place
that the guest can't handle.


> A similar approach was introduced in the SNP patches V1 and V2 for
> kernel page fault handling. The pushback around this convert to shared
> approach was largely focused around the idea that the kernel has all
> the information about which pages are shared vs private so it should
> be able to check shared status before write to pages. After V2 the
> patches were updated to not have a kernel page fault handler for RMP
> violations (other than dumping state during a panic). The current
> patches protect the host with new post_{map,unmap}_gfn() function that
> checks if a page is shared before mapping it, then locks the page
> shared until unmapped. Given the discussions on ‘[Part2,v5,39/45] KVM:
> SVM: Introduce ops for the post gfn map and unmap’ building a solution
> to do this is non trivial and adds new overheads to KVM. Additionally
> the current solution is local to the kernel. So a new ABI just now be
> created to allow the userspace VMM to access the kernel-side locks for
> this to work generically for the whole host. This is more complicated
> than this proposal and adding more lock holders seems like it could
> reduce performance further.

The locking is complicated.  But, I think it is necessary.  Once the
kernel can map private memory, it can access it in random spots.  It's
hard to make random kernel code recoverable from faults.