potential bug where mem_event_request_t gfn value == -1?

potential bug where mem_event_request_t gfn value == -1?

[Bob Jung]

Hi there,

I am running an adapted version of tools/tests/xen_access/xen_access.c  to get and respond to memory events to do some analysis.  Anyways, after registering for and receiving single stepping events in my ring buffer where the me_event_request_t's  reason field == MEM_EVENT_REASON_SINGLESTEP, I sometimes see that the gfn in the request is -1.

the guest is HVM windows 32 bit.  For 99.99% of the single step traps the req.gfn value is the correct guest frame number value, but every now and again I get a -1

Does anyone know why this could happen?  

Thanks,
-Bob

Re: potential bug where mem_event_request_t gfn value == -1?

[Tim Deegan]

At 20:49 +0000 on 06 Jan (1325882993), Bob Jung wrote:
> 
> Hi there,
> 
> I am running an adapted version of tools/tests/xen_access/xen_access.c  to get and respond to memory events to do some analysis.  Anyways, after registering for and receiving single stepping events in my ring buffer where the me_event_request_t's  reason field == MEM_EVENT_REASON_SINGLESTEP, I sometimes see that the gfn in the request is -1.
> 
> the guest is HVM windows 32 bit.  For 99.99% of the single step traps the req.gfn value is the correct guest frame number value, but every now and again I get a -1
> 
> Does anyone know why this could happen?  

If could happen if the VA in %rip at the time isn't mapped to anything
in the guest's pagetables (for example, if the guest is demand-paging
the program that's running).   You should be able to test this by looking
at the valie if %rip returned in the gla field.   There's some
pagetable-walking code in tools/libxc/xc_pagetab.c

Cheers,

Tim.