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

Re: [Xen-devel] Xen-4.5 HVMOP ABI issues

On 04/12/14 14:55, Jan Beulich wrote:
>>>> On 04.12.14 at 15:28, <andrew.cooper3@xxxxxxxxxx> wrote:
>> On 04/12/14 13:49, Jan Beulich wrote:
>>>>>> On 28.11.14 at 16:46, <andrew.cooper3@xxxxxxxxxx> wrote:
>>>> On 28/11/14 15:18, Jan Beulich wrote:
>>>>>>>> On 28.11.14 at 14:55, <andrew.cooper3@xxxxxxxxxx> wrote:
>>>>>> The problem is with continuations which reuse the upper bits of the
>>>>>> input register, not with this HVMOP_op_mask specifically; the
>>>>>> HVMOP_op_mask simply adds to an existing problem.  This is something
>>>>>> which needs considering urgently because, as you identify above, we have
>>>>>> already suffered an accidental ABI breakage with the mem-op widening.
>>>>> Since we can't retroactively fix the mem-op widening, I still don't see
>>>>> what's urgent here: As long as we don't change any of these masks,
>>>>> nothing bad is going to happen. Of course one thing to consider with
>>>>> this aspect in mind is whether to change the hvm-op or gnttab-op
>>>>> masks again _before_ 4.5 goes out, based on whether we feel they're
>>>>> wide enough for the (un)foreseeable future.
>>>> By urgent, I mean exactly this, while we have the ability to tweak the
>>>> masks.
>>> With no-one else voicing an opinion:
>>> For hvmop, the mask currently is 8 bits and we've got 22 ops defined.
>>> For gnttabop, the mask currently is 12 bits and we've got 12 ops defined.
>>> For the latter, we're fine even without further consideration. For the
>>> former, the two operations actively using the continuation encoding
>>> are tools-only ones. Since we're fine to alter the tools only interfaces,
>>> and since it was intended for the tools-only HVM-ops to be split off
>>> to a separate hypercall (e.g. hvmctl) anyway, the range restriction
>>> would then no longer be a problem. Plus, in the worst case we could
>>> always introduce yet another hypercall if we ran out of numbers.
>> Are you suggesting that we make a new hvmctl now and remove the hvmop
>> mask before 4.5?  If we ship 4.5 with the hvmop mask, we cannot
>> subsequently remove it even if all continuable hypercalls move to a
>> separate hypercall.
> Why? We certainly don't guarantee compatibility for undefined
> hypercalls to behave in a certain way.

A task is in the middle of a hypercall continuation.  The VM is migrated
to a newer Xen which has lost the op mask and gained a new hypercall
which would alias.

At this point, the task has transparently swapped hypercall subops,
through no fault of its own.

As I have said before, once an op mask has found its way into a released
version of Xen, it is irrevocably part of the ABI, and cannot change in
the future.


Xen-devel mailing list



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