Re: [PATCH v12 05/17] iommu: Add attach/detach_dev_pasid iommu interface

[Baolu Lu]

On 2022/8/26 22:37, Jason Gunthorpe wrote:
> On Fri, Aug 26, 2022 at 08:11:29PM +0800, Lu Baolu wrote:
>
> > + * iommu_get_domain_for_dev_pasid() - Retrieve domain for @pasid of @dev
> > + * @dev: the queried device
> > + * @pasid: the pasid of the device
> > + * @type: matched domain type, 0 for any match
> > + *
> > + * This is a variant of iommu_get_domain_for_dev(). It returns the existing
> > + * domain attached to pasid of a device. It's only for internal use of the
> > + * IOMMU subsystem.
>
> If it is only for external use then why is it exported?
>
> I would add something like:
>
>   Callers must hold a lock around this function, and both
>   iommu_attach/detach_dev_pasid() whenever a domain of type is being
>   manipulated. This API does not internally resolve races with
>   attach/detach.

Yes. Updated. This is what this API expected.

> > + * detaching from the device PASID.
> > + *
> > + * Return: attached domain on success, NULL otherwise.
> > + */
> > +struct iommu_domain *iommu_get_domain_for_dev_pasid(struct device *dev,
> > +						    ioasid_t pasid,
> > +						    unsigned int type)
> > +{
> > +	struct iommu_domain *domain;
> > +	struct iommu_group *group;
> > +
> > +	group = iommu_group_get(dev);
> > +	if (!group)
> > +		return NULL;
> > +	/*
> > +	 * The xarray protects its internal state with RCU. Hence the domain
> > +	 * obtained is either NULL or fully formed.
> > +	 */
>
> This has nothing to do with RCU
>
> xa_lock() is used to ensure that the domain pointer remains valid
> while we check the type since it blocks concurrent xa_erase().

With xa_lock() added, this comment is not needed anymore.

Best regards,
baolu