Re: [PATCH 4.19 17/32] iommu/vt-d: Dont queue_iova() if there is no flush queue

[Pavel Machek]

Hi!

> --- a/drivers/iommu/intel-iommu.c
> +++ b/drivers/iommu/intel-iommu.c
> @@ -3721,7 +3721,7 @@ static void intel_unmap(struct device *d
>  
>  	freelist = domain_unmap(domain, start_pfn, last_pfn);
>  
> -	if (intel_iommu_strict) {
> +	if (intel_iommu_strict || !has_iova_flush_queue(&domain->iovad)) {
>  		iommu_flush_iotlb_psi(iommu, domain, start_pfn,
>  				      nrpages, !freelist, 0);
>  		/* free iova */
> --- a/drivers/iommu/iova.c
> +++ b/drivers/iommu/iova.c
> @@ -65,9 +65,14 @@ init_iova_domain(struct iova_domain *iov
>  }
>  EXPORT_SYMBOL_GPL(init_iova_domain);
>  
> +bool has_iova_flush_queue(struct iova_domain *iovad)
> +{
> +	return !!iovad->fq;

Should this be READ_ONCE()?

> @@ -100,13 +106,17 @@ int init_iova_flush_queue(struct iova_do
>  	for_each_possible_cpu(cpu) {
>  		struct iova_fq *fq;
>  
> -		fq = per_cpu_ptr(iovad->fq, cpu);
> +		fq = per_cpu_ptr(queue, cpu);
>  		fq->head = 0;
>  		fq->tail = 0;
>  
>  		spin_lock_init(&fq->lock);
>  	}
>  
> +	smp_wmb();
> +
> +	iovad->fq = queue;
> +

Could we have a comment why the barrier is needed, and perhaps there
should be oposing smp_rmb() somewhere? Does this need to be
WRITE_ONCE() as it is racing against reader?

Thanks,
									Pavel

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Attachment: signature.asc
Description: Digital signature