Re: [PATCH RFC 1/5]Add some trace enties and define interface for tracing

[Avi Kivity <avi@qumranet.com>]

Liu, Eric E wrote:
>>
>> High order 32 bits of cr2 are lost.
>>
>>     
>
> May I use  KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
> (u32)((u64)cr2 >> 32), handler) to handle this?
> for 32bit gust, it traces some excess data, but after all for 64bit
> guest, we don't lost high order bits.
>
>   

Sure.

>   
>
>>>  		}
>>> -
>>> +		KVMTRACE_1D(INJ_VIRQ, vcpu, idtv_info_field, handler);
>>>
>>>       
>> Probably need a different marker than INJ_VIRQ, as this is on exit,
>> not entry.
>>
>>     
>
> Is the marker REDELIVER_EVT ok for this?
>
>   

Yes.

>>> @@ -2428,6 +2445,7 @@ void kvm_arch_exit(void)
>>>  int kvm_emulate_halt(struct kvm_vcpu *vcpu)
>>>  {
>>>  	++vcpu->stat.halt_exits;
>>> +	KVMTRACE_0D(HLT, vcpu, handler);
>>>  	if (irqchip_in_kernel(vcpu->kvm)) {
>>>  		vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
>>>  		kvm_vcpu_block(vcpu);
>>>
>>>       
>> Would be nice to have an UNHLT to see how long sleeps are.  But this
>> will probably be seen by the irq injection.
>>     
>
> I think from the cycles Between VMEXIT( caused by halt) and VMENTRY we
> can evaluate how long sleeps are. 
>   

Right.

We'll merge it with the understanding that the data format is not part 
of the ABI and will change between versions.  It will cause some 
headaches when people send us traces, but on the other hand will give us 
some flexibility.

-- 
Any sufficiently difficult bug is indistinguishable from a feature.


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Register now and save $200. Hurry, offer ends at 11:59 p.m., 
Monday, April 7! Use priority code J8TLD2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone