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

Re: [Xen-devel] [PATCH v4 1/2] x86/mm: factor out the code for shattering an l3 PTE


  • To: Jan Beulich <jbeulich@xxxxxxxx>
  • From: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
  • Date: Fri, 13 Dec 2019 14:58:13 +0000
  • Authentication-results: esa5.hc3370-68.iphmx.com; dkim=none (message not signed) header.i=none; spf=None smtp.pra=andrew.cooper3@xxxxxxxxxx; spf=Pass smtp.mailfrom=Andrew.Cooper3@xxxxxxxxxx; spf=None smtp.helo=postmaster@xxxxxxxxxxxxxxx
  • Autocrypt: addr=andrew.cooper3@xxxxxxxxxx; prefer-encrypt=mutual; keydata= mQINBFLhNn8BEADVhE+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+cWwARAQABtClBbmRyZXcgQ29v cGVyIDxhbmRyZXcuY29vcGVyM0BjaXRyaXguY29tPokCOgQTAQgAJAIbAwULCQgHAwUVCgkI 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 TQTBLzDKXok86LkCDQRS4TZ/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 uAJyDDKN99ziC0Wz5kcPyVD1HNf8bgaqGDzrv3TfYjwqayRFcMf7xJaL9xXedMcAEQEAAYkC HwQYAQgACQUCUuE2fwIbDAAKCRBlw/kGpdefoG4XEACD1Qf/er8EA7g23HMxYWd3FXHThrVQ HgiGdk5Yh632vjOm9L4sd/GCEACVQKjsu98e8o3ysitFlznEns5EAAXEbITrgKWXDDUWGYxd pnjj2u+GkVdsOAGk0kxczX6s+VRBhpbBI2PWnOsRJgU2n10PZ3mZD4Xu9kU2IXYmuW+e5KCA vTArRUdCrAtIa1k01sPipPPw6dfxx2e5asy21YOytzxuWFfJTGnVxZZSCyLUO83sh6OZhJkk b9rxL9wPmpN/t2IPaEKoAc0FTQZS36wAMOXkBh24PQ9gaLJvfPKpNzGD8XWR5HHF0NLIJhgg 4ZlEXQ2fVp3XrtocHqhu4UZR4koCijgB8sB7Tb0GCpwK+C4UePdFLfhKyRdSXuvY3AHJd4CP 4JzW0Bzq/WXY3XMOzUTYApGQpnUpdOmuQSfpV9MQO+/jo7r6yPbxT7CwRS5dcQPzUiuHLK9i nvjREdh84qycnx0/6dDroYhp0DFv4udxuAvt1h4wGwTPRQZerSm4xaYegEFusyhbZrI0U9tJ B8WrhBLXDiYlyJT6zOV2yZFuW47VrLsjYnHwn27hmxTC/7tvG3euCklmkn9Sl9IAKFu29RSo d5bD8kMSCYsTqtTfT6W4A3qHGvIDta3ptLYpIAOD2sY3GYq2nf3Bbzx81wZK14JdDDHUX2Rs 6+ahAA==
  • Cc: xen-devel@xxxxxxxxxxxxxxxxxxxx, jgrall@xxxxxxxxxx, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Wei Liu <wl@xxxxxxx>, Hongyan Xia <hongyxia@xxxxxxxxxx>
  • Delivery-date: Fri, 13 Dec 2019 14:58:39 +0000
  • Ironport-sdr: 4qbqjSzWM96TudEsyt/Vs4/4WPczfhlM3cp3v+lPVR5nsZWODjC5ScAxuLKcJB23uIPsw5tMYY zpFKt1Mhk3B4j+llHRfR2F/e8qDBSKV+DIr730B5w2Fm2yrnblGUNWhDFO3IUczD7GtnQX6EUV VIGx2uwy6vCeZnWJCzHIXUXE27wx/p067GocPMAZJHFqttHgDlZyht3dQG0PzlcKLYOOYnmGGo fCMFGPo3nP1qrXUweQqICQRUm/TBTYI/BnHwzcSXmKG8ZbYN5OpsFQN2g/88aMZZxPxoIUmpSN jsg=
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
  • Openpgp: preference=signencrypt

On 13/12/2019 14:36, Jan Beulich wrote:
> On 13.12.2019 15:19, Andrew Cooper wrote:
>> On 12/12/2019 12:46, Hongyan Xia wrote:
>>> diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
>>> index 7d4dd80a85..8def4fb8d8 100644
>>> --- a/xen/arch/x86/mm.c
>>> +++ b/xen/arch/x86/mm.c
>>> @@ -5151,6 +5151,52 @@ l1_pgentry_t *virt_to_xen_l1e(unsigned long v)
>>>                           flush_area_local((const void *)v, f) : \
>>>                           flush_area_all((const void *)v, f))
>>>  
>>> +/* Shatter an l3 entry and populate l2. If virt is passed in, also do 
>>> flush. */
>>> +static bool shatter_l3e(l3_pgentry_t *pl3e, unsigned long virt, bool 
>>> locking)
>>> +{
>>> +    unsigned int i;
>>> +    l3_pgentry_t ol3e = *pl3e;
>>> +    l2_pgentry_t l2e = l2e_from_intpte(l3e_get_intpte(ol3e));
>>> +    l2_pgentry_t *l2t = alloc_xen_pagetable();
>>> +
>>> +    if ( !l2t )
>>> +        return false;
>>> +
>>> +    for ( i = 0; i < L2_PAGETABLE_ENTRIES; i++ )
>>> +    {
>>> +        l2e_write(l2t + i, l2e);
>>> +        l2e = l2e_from_intpte(
>>> +                  l2e_get_intpte(l2e) + (PAGE_SIZE << PAGETABLE_ORDER));
>>> +    }
>>> +
>>> +    if ( locking )
>>> +        spin_lock(&map_pgdir_lock);
>>> +    if ( (l3e_get_flags(*pl3e) & _PAGE_PRESENT) &&
>>> +         (l3e_get_flags(*pl3e) & _PAGE_PSE) )
>>> +    {
>>> +        l3e_write_atomic(pl3e,
>>> +            l3e_from_paddr(virt_to_maddr(l2t), __PAGE_HYPERVISOR));
>>> +        l2t = NULL;
>>> +    }
>>> +    if ( locking )
>>> +        spin_unlock(&map_pgdir_lock);
>>> +
>>> +    if ( virt )
>>> +    {
>>> +        unsigned int flush_flags =
>>> +            FLUSH_TLB | FLUSH_ORDER(2 * PAGETABLE_ORDER);
>>> +
>>> +        if ( l3e_get_flags(ol3e) & _PAGE_GLOBAL )
>>> +            flush_flags |= FLUSH_TLB_GLOBAL;
>> Another problematic use of ol3e which is racy on conflict.  You need to
>> strictly use the content of *pl3e from within the locked region.
> But this isn't a problem introduced here, i.e. fixing of it doesn't
> strictly fall under "re-factor". (I'm certainly not opposed to
> getting this right at the same time.)

It is brand new code which is racy.  Its either not necessary, or an
XSA-in-waiting.  (And not necessary, AFAICT).

>
>> However, why have you moved the flushing in here?  Only one of the two
>> callers actually wanted it, and even then I'm not totally sure it is
>> necessary.
>>
>> Both callers operate on an arbitrary range of addresses, and for
>> anything other than a singleton update, will want to issue a single
>> flush at the end, rather than a spate of flushes for sub-areas.
>>
>> (Although someone really please check my reasoning here for the
>> map_pages_to_xen() case which currently does have sub-area flushing.)
>>
>> Either the flush wants dropping (and best via a prereq patch altering
>> map_pages_to_xen()), or you need to cache ol3e in the locked region with
>> ACCESS_ONCE() or equivalent.
> Well, at best replacing by a single one at the end, but I guess
> the current piecemeal behavior is to cope with error paths (see
> Julien's report against modify_xen_mappings(), where it's
> exactly the other way around). Considering especially speculative
> accesses I think it isn't the worst idea to keep the window small
> between shatter and flush (short of us doing a proper break-then-
> make sequence).

Every sub-flush is a broadcast IPI, which is a scalability concern, and
at least needs considering.

x86 is designed not to need BBM, although the BBM sequence can be
helpful at times to simplify other reasoning.  It might actually be
necessary in some SMP cases.

Speculation can bite you at any point, including the very next
instruction.  The logic is either correct, or not correct, and the
distance between the PTE update and the flush is only relevant when it
comes to the scarcity of the incorrect case manifesting in a noticeable way.

Fundamentally, we either need BBM, or the flush is safe to defer to the
end.  Everything in-between is racy, and dropping the sub-flushes would
make any incorrect cases more likely to manifest.

~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®.