Re: [PATCH 3/5] xen/vm_event: Support for guest-requested events

[Razvan Cojocaru <rcojocaru@bitdefender.com>]

On 05/07/2015 08:05 PM, Tamas K Lengyel wrote:
> Hi Razvan,
> I guess I don't really see into the setup you envision using this
> event-type with, but wouldn't it be possible to just use libxenvchan?

Hello Tamas,

This patch evolved from "xen: Support for VMCALL mem_events":

http://lists.xen.org/archives/html/xen-devel/2014-07/msg00313.html

that I've sent in a previous series. Basically, it allows the guest to
trigger a vm_event that contains some of the guest's state. In the
current incarnation of the patch, the state consists of the x86
registers that have been filled in prior to sending out the vm_event.
So, just a way for the guest itself to be able to send out a generic
vm_event to any interested consumer. We thought that might be
interesting for everyone, since the event can be used for almost
anything and is not memory-introspection (or any particular thing) specific.

The main difference between the initial patch and this one (besides the
vm_event work that has taken place since) is that this is now a proper
hypercall and the magic numbers are gone. I have to thank Andrew Cooper
for the suggestion to do it this way.

To answer your question, I haven't really looked closely at libxenvchan
before, but our use case at least can't use it (we're injecting some
code in the guest on the fly). Am I wrong in assuming libxenvchan also
requires two PV domains talking to each other? The guests we're
monitoring are HVM guests, and _they_ are supposed to issue the
VMCALL-triggered events.


Thanks,
Razvan