Re: [PATCH/RFC] KVM: track pid for VCPU only on KVM_RUN ioctl

[Paolo Bonzini <pbonzini@redhat.com>]

Il 19/08/2014 11:47, Christian Borntraeger ha scritto:
> On 19/08/14 11:27, Paolo Bonzini wrote:
>> Il 19/08/2014 10:38, Christian Borntraeger ha scritto:
>>>>> The patch may be okay, but I'm worried that it might be
>>>>> hiding a bug in QEMU.
>>> On s390 we call "KVM_S390_INITIAL_RESET" from several reset
>>> functions, e.g. during CPU creation. This is the first hickup and
>>> the pid now points to the main thread.
>> 
>> Any reason to have a special ioctl instead of
>> SET_REGS/SET_ONE_REG/... (via kvm_cpu_synchronize_state, which does
>> the ioctls in the VCPU thread)?
> 
> Historical reasons mostly. Older kernel miss several interfaces to
> bring the CPU in a defined state (pending interrupts, cpu state, some
> registers...)
> 
> Good news is that we are working on getting rid of it: cpu states are
> now available as far as I can see, only local interrupt flushing is
> missing.This needs some more work on our side.  So in some month we
> probably will have a QEMU version that does not need to call this any
> more. For todays QEMU this patch help though.

Just by the sound of it, interrupt flushing seems dangerous to do in a
way that could be concurrent with KVM_RUN...

Paolo