Re: [PATCH v2 2/2] iommu/vt-d: Use device rbtree in iopf reporting path

[Jason Gunthorpe]

On Tue, Feb 20, 2024 at 02:59:39PM +0800, Lu Baolu wrote:
> diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
> index acfe27bd3448..6743fe6c7a36 100644
> --- a/drivers/iommu/intel/iommu.c
> +++ b/drivers/iommu/intel/iommu.c
> @@ -4430,8 +4430,11 @@ static struct iommu_device *intel_iommu_probe_device(struct device *dev)
>  static void intel_iommu_release_device(struct device *dev)
>  {
>  	struct device_domain_info *info = dev_iommu_priv_get(dev);
> +	struct intel_iommu *iommu = info->iommu;
>  
> +	mutex_lock(&iommu->iopf_lock);
>  	device_rbtree_remove(info);
> +	mutex_unlock(&iommu->iopf_lock);

This seems like a pretty reasonable solution, maybe someday it can
become lockless.. This is a fast path right?

> @@ -691,21 +691,22 @@ static irqreturn_t prq_event_thread(int irq, void *d)
>  		if (unlikely(req->lpig && !req->rd_req && !req->wr_req))
>  			goto prq_advance;
>  
> -		pdev = pci_get_domain_bus_and_slot(iommu->segment,
> -						   PCI_BUS_NUM(req->rid),
> -						   req->rid & 0xff);
>  		/*
>  		 * If prq is to be handled outside iommu driver via receiver of
>  		 * the fault notifiers, we skip the page response here.
>  		 */
> -		if (!pdev)
> +		mutex_lock(&iommu->iopf_lock);
> +		dev = device_rbtree_find(iommu, req->rid);
> +		if (!dev) {
> +			mutex_unlock(&iommu->iopf_lock);
>  			goto bad_req;
> +		}

Though now we have a mutex and a spinlock covering the same data
structure.. It could be optimized some more.

But maybe we should leave micro optimization aside for now.

Reviewed-by: Jason Gunthorpe <jgg@xxxxxxxxxx>

Jason