Re: [Xen-devel] [PATCH v3] xen: introduce VCPUOP_register_runstate_phys_memory_area hypercall

[Julien Grall <julien.grall@xxxxxxx>]

Hi Andrii,

On 14/06/2019 16:11, Andrii Anisov wrote:
> On 14.06.19 17:39, Julien Grall wrote:
> > Why? What are the benefits for a guest to use the two interface together?
> I do not say the guest has to use both interfaces simultaneously. It is 
> logically odd, doing so will only reflect in increasing of hypervisor overhead.
> But such an implementation will have a simpler code, which expected to be (a 
> bit) faster. > So the code simplicity would be a benefit for us. Lower hypervisor overhead is a
> benefit for sane guests, which use only one interface.
I hope you are aware that speaking about speed here is quite irrelevant. The 
difference would be clear lost in the noise of the rest of the context switch.
But, if you allow something, then most likely someone will use it. However, you 
have to differentiate implementation vs documentation.
In this case, I don't think the implementation should dictate what is going to 
be exposed.
If you document that it can't happen, then you have room to forbid the mix in 
the future (assuming this can't be done now).
In other word, the more lax is the interface, the more difficult it is tighten 
in the future.
I am not going to push for an implementation that forbid the mix. But I am 
strongly going to push for any documentation of the expected interaction. So we 
don't make our life miserable later on.
> BTW, dropping the old interface implementation will be much easier in future if 
> it will not clash with the new one.
I am afraid we will never be able to remove the old interface.

> > After all they have exactly the same data...
> Yes, but normal guests should use only one interface.
See above.

Cheers,

--
Julien Grall

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