Re: [PATCH RFC V5 2/6] kvm hypervisor : Add a hypercall to KVM hypervisor to support pv-ticketlocks

[Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>]

Sorry for late reply,
was on vacation for a week (without IMAP access :( )

On 04/12/2012 05:36 AM, Marcelo Tosatti wrote:
> On Fri, Mar 23, 2012 at 01:37:04PM +0530, Raghavendra K T wrote:
>> From: Srivatsa Vaddagiri<vatsa@linux.vnet.ibm.com>
[snip]
>> @@ -1567,6 +1568,9 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
>>   		prepare_to_wait(&vcpu->wq,&wait, TASK_INTERRUPTIBLE);
>>
>>   		if (kvm_arch_vcpu_runnable(vcpu)) {
>> +			vcpu->pv_unhalted = 0;
>> +			/* preventing reordering should be enough here */
>> +			barrier();
>
> Is it always OK to erase the notification, even in case an unrelated
> event such as interrupt was the source of wakeup?


Erasing notification is not good, But I think in this case,

kvm_make_request(KVM_REQ_UNHALT, vcpu);

below this would take care of the rest.

>
> It would be easier to verify that notifications are not lost with atomic
>
> test_and_clear(pv_unhalted).

true, I 'll verify that (with pv_unhalt as atomic variable). my heart
says  current code is just fine, since we are about to unblock.

>
> Also x86 specific code should remain in arch/x86/kvm/
>

I agree. 'll have clear function in arch/x86/kvm and add stub to rest
of the archs

>