[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH v2] x86/mm: Suppresses vm_events caused by page-walks


  • To: Jan Beulich <JBeulich@xxxxxxxx>, <aisaila@xxxxxxxxxxxxxxx>
  • From: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
  • Date: Tue, 18 Sep 2018 19:20:46 +0100
  • Autocrypt: addr=andrew.cooper3@xxxxxxxxxx; prefer-encrypt=mutual; keydata= xsFNBFLhNn8BEADVhE+Hb8i0GV6mihnnr/uiQQdPF8kUoFzCOPXkf7jQ5sLYeJa0cQi6Penp VtiFYznTairnVsN5J+ujSTIb+OlMSJUWV4opS7WVNnxHbFTPYZVQ3erv7NKc2iVizCRZ2Kxn srM1oPXWRic8BIAdYOKOloF2300SL/bIpeD+x7h3w9B/qez7nOin5NzkxgFoaUeIal12pXSR Q354FKFoy6Vh96gc4VRqte3jw8mPuJQpfws+Pb+swvSf/i1q1+1I4jsRQQh2m6OTADHIqg2E ofTYAEh7R5HfPx0EXoEDMdRjOeKn8+vvkAwhviWXTHlG3R1QkbE5M/oywnZ83udJmi+lxjJ5 YhQ5IzomvJ16H0Bq+TLyVLO/VRksp1VR9HxCzItLNCS8PdpYYz5TC204ViycobYU65WMpzWe LFAGn8jSS25XIpqv0Y9k87dLbctKKA14Ifw2kq5OIVu2FuX+3i446JOa2vpCI9GcjCzi3oHV e00bzYiHMIl0FICrNJU0Kjho8pdo0m2uxkn6SYEpogAy9pnatUlO+erL4LqFUO7GXSdBRbw5 gNt25XTLdSFuZtMxkY3tq8MFss5QnjhehCVPEpE6y9ZjI4XB8ad1G4oBHVGK5LMsvg22PfMJ ISWFSHoF/B5+lHkCKWkFxZ0gZn33ju5n6/FOdEx4B8cMJt+cWwARAQABzSlBbmRyZXcgQ29v cGVyIDxhbmRyZXcuY29vcGVyM0BjaXRyaXguY29tPsLBegQTAQgAJAIbAwULCQgHAwUVCgkI CwUWAgMBAAIeAQIXgAUCWKD95wIZAQAKCRBlw/kGpdefoHbdD/9AIoR3k6fKl+RFiFpyAhvO 59ttDFI7nIAnlYngev2XUR3acFElJATHSDO0ju+hqWqAb8kVijXLops0gOfqt3VPZq9cuHlh IMDquatGLzAadfFx2eQYIYT+FYuMoPZy/aTUazmJIDVxP7L383grjIkn+7tAv+qeDfE+txL4 SAm1UHNvmdfgL2/lcmL3xRh7sub3nJilM93RWX1Pe5LBSDXO45uzCGEdst6uSlzYR/MEr+5Z JQQ32JV64zwvf/aKaagSQSQMYNX9JFgfZ3TKWC1KJQbX5ssoX/5hNLqxMcZV3TN7kU8I3kjK mPec9+1nECOjjJSO/h4P0sBZyIUGfguwzhEeGf4sMCuSEM4xjCnwiBwftR17sr0spYcOpqET ZGcAmyYcNjy6CYadNCnfR40vhhWuCfNCBzWnUW0lFoo12wb0YnzoOLjvfD6OL3JjIUJNOmJy RCsJ5IA/Iz33RhSVRmROu+TztwuThClw63g7+hoyewv7BemKyuU6FTVhjjW+XUWmS/FzknSi dAG+insr0746cTPpSkGl3KAXeWDGJzve7/SBBfyznWCMGaf8E2P1oOdIZRxHgWj0zNr1+ooF /PzgLPiCI4OMUttTlEKChgbUTQ+5o0P080JojqfXwbPAyumbaYcQNiH1/xYbJdOFSiBv9rpt TQTBLzDKXok86M7BTQRS4TZ/ARAAkgqudHsp+hd82UVkvgnlqZjzz2vyrYfz7bkPtXaGb9H4 Rfo7mQsEQavEBdWWjbga6eMnDqtu+FC+qeTGYebToxEyp2lKDSoAsvt8w82tIlP/EbmRbDVn 7bhjBlfRcFjVYw8uVDPptT0TV47vpoCVkTwcyb6OltJrvg/QzV9f07DJswuda1JH3/qvYu0p vjPnYvCq4NsqY2XSdAJ02HrdYPFtNyPEntu1n1KK+gJrstjtw7KsZ4ygXYrsm/oCBiVW/OgU g/XIlGErkrxe4vQvJyVwg6YH653YTX5hLLUEL1NS4TCo47RP+wi6y+TnuAL36UtK/uFyEuPy wwrDVcC4cIFhYSfsO0BumEI65yu7a8aHbGfq2lW251UcoU48Z27ZUUZd2Dr6O/n8poQHbaTd 6bJJSjzGGHZVbRP9UQ3lkmkmc0+XCHmj5WhwNNYjgbbmML7y0fsJT5RgvefAIFfHBg7fTY/i kBEimoUsTEQz+N4hbKwo1hULfVxDJStE4sbPhjbsPCrlXf6W9CxSyQ0qmZ2bXsLQYRj2xqd1 bpA+1o1j2N4/au1R/uSiUFjewJdT/LX1EklKDcQwpk06Af/N7VZtSfEJeRV04unbsKVXWZAk uAJyDDKN99ziC0Wz5kcPyVD1HNf8bgaqGDzrv3TfYjwqayRFcMf7xJaL9xXedMcAEQEAAcLB XwQYAQgACQUCUuE2fwIbDAAKCRBlw/kGpdefoG4XEACD1Qf/er8EA7g23HMxYWd3FXHThrVQ HgiGdk5Yh632vjOm9L4sd/GCEACVQKjsu98e8o3ysitFlznEns5EAAXEbITrgKWXDDUWGYxd pnjj2u+GkVdsOAGk0kxczX6s+VRBhpbBI2PWnOsRJgU2n10PZ3mZD4Xu9kU2IXYmuW+e5KCA vTArRUdCrAtIa1k01sPipPPw6dfxx2e5asy21YOytzxuWFfJTGnVxZZSCyLUO83sh6OZhJkk b9rxL9wPmpN/t2IPaEKoAc0FTQZS36wAMOXkBh24PQ9gaLJvfPKpNzGD8XWR5HHF0NLIJhgg 4ZlEXQ2fVp3XrtocHqhu4UZR4koCijgB8sB7Tb0GCpwK+C4UePdFLfhKyRdSXuvY3AHJd4CP 4JzW0Bzq/WXY3XMOzUTYApGQpnUpdOmuQSfpV9MQO+/jo7r6yPbxT7CwRS5dcQPzUiuHLK9i nvjREdh84qycnx0/6dDroYhp0DFv4udxuAvt1h4wGwTPRQZerSm4xaYegEFusyhbZrI0U9tJ B8WrhBLXDiYlyJT6zOV2yZFuW47VrLsjYnHwn27hmxTC/7tvG3euCklmkn9Sl9IAKFu29RSo d5bD8kMSCYsTqtTfT6W4A3qHGvIDta3ptLYpIAOD2sY3GYq2nf3Bbzx81wZK14JdDDHUX2Rs 6+ahAA==
  • Cc: Tamas K Lengyel <tamas@xxxxxxxxxxxxx>, Wei Liu <wei.liu2@xxxxxxxxxx>, Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>, George Dunlap <George.Dunlap@xxxxxxxxxxxxx>, Tim Deegan <tim@xxxxxxx>, xen-devel@xxxxxxxxxxxxx
  • Delivery-date: Tue, 18 Sep 2018 18:25:57 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
  • Openpgp: preference=signencrypt

On 18/09/18 11:17, Jan Beulich wrote:
>>>> On 18.09.18 at 11:47, <aisaila@xxxxxxxxxxxxxxx> wrote:
>> On Thu, 2018-09-13 at 08:17 -0600, Jan Beulich wrote:
>>>>>> On 12.09.18 at 11:47, <aisaila@xxxxxxxxxxxxxxx> wrote:
>>>> The original version of the patch emulated the current instruction
>>>> (which, as a side-effect, emulated the page-walk as well), however
>>>> we
>>>> need finer-grained control. We want to emulate the page-walk, but
>>>> still
>>>> get an EPT violation event if the current instruction would trigger
>>>> one.
>>>> This patch performs just the page-walk emulation.
>>> Rather than making this basically a revision log, could you please
>>> focus
>>> on what you actually want to achieve?
>>>
>>> As to the title: "Suppress ..." please.
>>>
>>>> @@ -149,6 +151,10 @@ guest_walk_tables(struct vcpu *v, struct
>>>> p2m_domain *p2m,
>>>>      ar_and &= gflags;
>>>>      ar_or  |= gflags;
>>>>  
>>>> +    if ( set_ad && set_ad_bits(&l4p[guest_l4_table_offset(va)].l4,
>>>> +                               &gw->l4e.l4, false) )
>>>> +        accessed = true;
>>> It is in particular this seemingly odd (and redundant with what's
>>> done
>>> later in the function) which needs thorough explanation.
>> On this patch I've followed Andrew Cooper's suggestion on how to set
>> A/D Bits:
>>
>> "While walking down the levels, set any missing A bits and remember if we
>> set any.  If we set A bits, consider ourselves complete and exit back to
>> the guest.  If no A bits were set, and the access was a write (which we
>> know from the EPT violation information), then set the leaf D bit."
>>
>> If I misunderstood the comment please clarify.
> It doesn't look to me as if you misunderstood anything, but only Andrew
> can say for sure. However, none of this was in the description of your
> patch (neither as part of the description, nor as code comment), and I
> think you'd even have to greatly extend on this in order to explain to
> everyone why the resulting behavior is still architecturally correct. In no
> case should you assume anyone reading your patch (now or in the
> future) has participated in the earlier discussion.

The problem we have is that, while we know the EPT Violation was for a
write of an A or D bit to a write-protected guest pagetable, we don't
know if it was the A or the D bit which was attempting to be set.

Furthermore (without emulating the instruction, which is what we are
trying to avoid), we can't reconstruct the access.

Access bits are only written if they were missing before, but may be set
speculatively.  Dirty bits are only set when a write is retired.  From a
practical point of view, the pipeline sets A and D bits as separate actions.

Following this logic (and assuming for now a single vcpu), if we get a
GPT EPT Violation, and there are missing access bits on the walk, then
the fault is definitely from setting an access bit.  Set all access bits
and call it done.  If we get a GPT EPT Violation and all access bits
were set, then it was definitely from setting the Dirty bit.

For multi-vcpu scenarios, things get racy.  Setting all the Access bits
is safe because its a speculative action, but a speculatively load on
one vcpu can race with a write (to a read-only mapping) on the other
vcpu, and would trick this algorithm into setting the dirty bit when the
write would have faulted (and not set the dirty bit).

Do we have numbers on how many the GPT EPT Violations are for (only)
access sets, and how many are for dirty tsets?  Would the first half of
the algorithm (which is definitely not racy) still be a net perf win?

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.