Re: [PATCH] KVM: VMX: Do not overwrite vcpu->srcu_idx in vmx_vcpu_reset

[Paolo Bonzini <pbonzini@redhat.com>]

Il 14/03/2013 16:11, Gleb Natapov ha scritto:
> On Thu, Mar 14, 2013 at 04:08:42PM +0100, Paolo Bonzini wrote:
>> Il 14/03/2013 16:03, Jan Kiszka ha scritto:
>>>>> vcpu->srcu_idx = srcu_read_lock()
>>>>>   idx = srcu_read_lock(&vcpu->kvm->srcu); 
>>>>>    srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
>>>>>    vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
>>>>>   srcu_read_unlock(&vcpu->kvm->srcu, idx);
>>>>> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
>>>>>
>>>>> Not sure this is valid.
>>> Grmbl, likely not.
>>
>> It might be.
>>
>> Isn't it the same as two different CPUs doing
>>
>> CPU 1                                                 CPU 2
>> ------------------------------------------------------------------------------------------------
>>
>> vcpu->srcu_idx = srcu_read_lock()
>>                                                       idx = srcu_read_lock(&vcpu->kvm->srcu); 
>> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
>> vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
>>                                                       srcu_read_unlock(&vcpu->kvm->srcu, idx);
>> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
>>
>> ?
>>
> Srcu may have per cpu state. We can always ask Paul.

There is per-CPU state but it is only used as an optimization.
synchronize_srcu only uses the sum of all values.

In fact, SRCU critical sections are preemptable so there's not even a
guarantee that srcu_read_lock() and srcu_read_unlock() run on the same CPU.

Paolo