Re: [patch 2/4] KVM: move coalesced_mmio locking to its own device

[Avi Kivity <avi@redhat.com>]

Marcelo Tosatti wrote:
> Get rid of kvm->lock dependency on coalesced_mmio methods. Use an 
> atomic variable instead to guarantee only one vcpu is batching
> data into the ring at a given time.
>
> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
>
> Index: kvm-irqlock/virt/kvm/coalesced_mmio.c
> ===================================================================
> --- kvm-irqlock.orig/virt/kvm/coalesced_mmio.c
> +++ kvm-irqlock/virt/kvm/coalesced_mmio.c
> @@ -26,9 +26,12 @@ static int coalesced_mmio_in_range(struc
>  	if (!is_write)
>  		return 0;
>  
> -	/* kvm->lock is taken by the caller and must be not released before
> -         * dev.read/write
> -         */
> +	/*
> + 	 * Some other vcpu might be batching data into the ring,
> + 	 * fallback to userspace. Ordering not our problem.
> + 	 */
> +	if (!atomic_add_unless(&dev->in_use, 1, 1))
> +		return 0;
>  
>   

Ordering with simultaneous writes is indeed not our problem, but the 
ring may contain ordered writes (even by the same vcpu!)

Suggest our own lock here.  in_use is basically a homemade lock, better 
to use the factory made ones which come with a warranty.


-- 
error compiling committee.c: too many arguments to function