Re: [Xen-devel] [PATCH] common/gnttab: Introduce command line feature controls

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 25/08/17 10:57, Jan Beulich wrote:
>>>> On 24.08.17 at 17:16, <andrew.cooper3@xxxxxxxxxx> wrote:
>> On 24/08/17 16:01, Juergen Gross wrote:
>>> On 24/08/17 16:50, Andrew Cooper wrote:
>>>> --- a/docs/misc/xen-command-line.markdown
>>>> +++ b/docs/misc/xen-command-line.markdown
>>>> @@ -868,6 +868,19 @@ Controls EPT related features.
>>>>  
>>>>  Specify which console gdbstub should use. See **console**.
>>>>  
>>>> +### gnttab
>>>> +> `= List of [ max_ver:<integer>, transitive ]`
>>>> +
>>>> +> Default: `gnttab=max_ver:2,transitive`
>>>> +
>>>> +Control various aspects of the grant table behaviour available to guests.
>>>> +
>>>> +* `max_ver` Select the maximum grant table version to offer to guests.  
>>>> Valid
>>>> +version are 1 and 2.
>>>> +* `transitive` Permit or disallow the use of transitive grants.  Note 
>>>> that the
>>>> +use of grant table v2 without transitive grants is an ABI breakage from 
>>>> the
>>>> +guests point of view.
>>> So shouldn't there be a way for the guest to query the support of
>>> transient grants?
>> Ideally yes, but how do you suggest doing this in a compatible way?
>>
>> All Xen downstreams which haven't backported the eventual transitive
>> fixes will have this clobber in place, without any query-ability.
> That workaround should not be used as an argument to not
> provide a way to query the capability. It was put in place knowing
> that it would cause problems for (hypothetical) guests using
> transitive grants.

I am not objecting to introducing a mechanism if a suitable one can be
found.

However, the heritage of XSA-226 is a valid reason to not block this
patch because a mechanism isn't present.

>
> I'm not sure Jürgen's ELF note suggestion would be very useful
> though: I don't see how Xen knowing a guest kernel can deal with
> the situation would change anything - I don't think we should
> fail the loading of a kernel without such a note when transitive
> grants are disabled, not the least because we know of no kernels
> using them, and hence we'd pointlessly prevent the use of older
> kernels in such a case.
>
> What about a negative XENFEAT_*? New code could query it,
> and existing code is hosed anyway if run on such a system.

Better yet, how about combining it with Juergens "xen: add new hypercall
to get grant table limits"?

We could have a features_available bitmap along with other gnttab
related maxima.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel