Re: [PATCH 9/9] iommu: Use fault cookie to store iopf_param

[Baolu Lu]

On 2023/7/11 14:26, Tian, Kevin wrote:
> > From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > Sent: Tuesday, July 11, 2023 9:07 AM
> >
> > Remove the static iopf_param pointer from struct iommu_fault_param to
> > save memory.
>
> why is there memory saving? you replace a single pointer with a xarray now...

iopf_param is duplicate with the fault cookie. So replace it with the
fault cookie to remove duplication and save memory.

> > @@ -303,16 +303,27 @@ int iopf_queue_add_device(struct iopf_queue
> > *queue, struct device *dev)
> >
> >   	mutex_lock(&queue->lock);
> >   	mutex_lock(&param->lock);
> > -	if (!param->iopf_param) {
> > -		list_add(&iopf_param->queue_list, &queue->devices);
> > -		param->iopf_param = iopf_param;
> > -		ret = 0;
> > +	curr = iommu_set_device_fault_cookie(dev, 0, iopf_param);
> > +	if (IS_ERR(curr)) {
> > +		ret = PTR_ERR(curr);
> > +		goto err_free;
> >   	}
>
> So although the new xarray is called a per-pasid storage, here only
> slot#0 is used for sva which includes a list containing partial req's
> for many pasid's. It doesn't sound clean...

Just to make it generic so that IOMMUFD can also use it. IOMMUFD will
use it to store the per-{device, pasid} object id (and possibly other
data) so that it can be quickly retrieved in the critical fault
delivering patch.

Best regards,
baolu