Re: Passing data from hypervisor to dom0 application

[Mats Petersson <mats@planetcatfish.com>]

At 08:16 14/07/2007, Abhinav Srivastava wrote:

>Hi there,
>
>I am working with Xen 3.0.3 in para-virtualized mode. I am 
>intercepting guest virtual machines hypercalls inside the the 
>hypervisor and trying to pass this data to dom0 monitoring 
>application. To pass data, I shared a page between dom0 and 
>hypervisor using alloc_hypertrace_buf function and used memcpy to 
>write data on the shared page. To inform userspace application about 
>the data, i used virq.
>
>The problem that i am facing in this whole process is since 
>hypercall data is coming so fast userspace application is not able 
>to show all the hypercalls.
>I am not able to think of exact reason of this behavior. Is it 
>because of virq's which are sent by Xen? or Xen is sending so fast 
>that userspace application is not able to read it? or Is it some 
>pending interrupt problem?


I don't know what the problem with this approach is, but two things 
come to mind:
1. Using more than one buffer and/or a larger buffer would make the 
overhead of the user-mode application smaller, which may help. More 
than one buffer would help in the respect that hypervisor can go on 
filling the next buffer after Dom0 app is starting it's copy, which 
may help to overcome any latency problems.
2. Isn't xentrace with suitable mask able to do exactly what you 
want? [I have only used xentrace to follow HVM type calls, but I 
expect that it's able to trace hypercalls too]. If it's not able to 
do that, you may want to consider adding some tracing to the 
hypercalls using the xentrace method, rather than inventing your own 
method of tracing - unless there is some major reason for NOT using 
Xentrace of course.

--
Mats