Re: [PATCH RFC 5/9] xen: Support for VMCALL mem_events

[Razvan Cojocaru <rcojocaru@bitdefender.com>]

On 07/02/2014 07:11 PM, Jan Beulich wrote:
>>>> On 02.07.14 at 17:54, <rcojocaru@bitdefender.com> wrote:
>>>> --- a/xen/arch/x86/hvm/vmx/vmx.c
>>>> +++ b/xen/arch/x86/hvm/vmx/vmx.c
>>>> @@ -2880,8 +2880,21 @@ void vmx_vmexit_handler(struct cpu_user_regs *regs)
>>>>      case EXIT_REASON_VMCALL:
>>>>      {
>>>>          int rc;
>>>> +        unsigned long eax = regs->eax;
>>>> +
>>>>          HVMTRACE_1D(VMMCALL, regs->eax);
>>>> -        rc = hvm_do_hypercall(regs);
>>>> +
>>>> +        if ( regs->eax != 0x494e5452 ) /* Introcore magic */
>>>
>>> Urgh?!
>>
>> The magic constant is INTR, and it's used to differentiate between
>> "regular" and induced VMCALLs. Our application sets EAX up like that to
>> tell the situations apart.
> 
> But that needs (a) a #define and (b) an explanation.
> 
> Plus - what keeps code outside of your app to invoke this?

Unfortunately, nothing. It can be set outside our application.

> 
>>>> --- a/xen/include/public/hvm/params.h
>>>> +++ b/xen/include/public/hvm/params.h
>>>> @@ -148,6 +148,8 @@
>>>>  #define HVM_PARAM_IOREQ_SERVER_PFN 32
>>>>  #define HVM_PARAM_NR_IOREQ_SERVER_PAGES 33
>>>>  
>>>> -#define HVM_NR_PARAMS          34
>>>> +#define HVM_PARAM_MEMORY_EVENT_VMCALL 34
>>>
>>> So why does this (used only as an argument to
>>> hvm_memory_event_traps()) need to be settable? I guess the patch
>>> description is just too brief.
>>
>> Settable?
> 
> You must have a reason to make this a HVM param. That reason is
> what I'm asking for.

I see. I want to be able to enable / disable this type of events. I.e.:

if (flags & ENABLE_VMCALL)
    xc_set_hvm_param(xci, domain, HVM_PARAM_MEMORY_EVENT_VMCALL,
                     HVMPME_mode_sync);

from the application, via libxc.


Thanks,
Razvan Cojocaru