Re: [PATCH v9 3/6] iommufd: Add iommufd_access_change_ioas helper

[Jason Gunthorpe]

On Thu, Jul 27, 2023 at 12:23:08AM -0700, Nicolin Chen wrote:
> The complication of the mutex and refcount will be amplified after we
> introduce the replace support for access. So, add a preparatory change
> of a constitutive helper iommufd_access_change_ioas(), to take care of
> the existing iommufd_access_attach() and iommufd_access_detach().
> 
> Also, update the unprotect routine in iommufd_access_destroy_object()
> to calling the new iommufd_access_change_ioas() helper.
> 
> Suggested-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> Signed-off-by: Nicolin Chen <nicolinc@xxxxxxxxxx>
> ---
>  drivers/iommu/iommufd/device.c | 112 ++++++++++++++++++++-------------
>  1 file changed, 69 insertions(+), 43 deletions(-)
> 
> diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
> index 7a3e8660b902..d9680a247e1c 100644
> --- a/drivers/iommu/iommufd/device.c
> +++ b/drivers/iommu/iommufd/device.c
> @@ -684,17 +684,69 @@ void iommufd_device_detach(struct iommufd_device *idev)
>  }
>  EXPORT_SYMBOL_NS_GPL(iommufd_device_detach, IOMMUFD);
>  
> +static int iommufd_access_change_ioas(struct iommufd_access *access,
> +				      struct iommufd_ioas *new_ioas)
> +{
> +	u32 iopt_access_list_id = access->iopt_access_list_id;
> +	struct iommufd_ioas *cur_ioas = access->ioas;
> +	int rc;
> +
> +	lockdep_assert_held(&access->ioas_lock);
> +
> +	/* We are racing with a concurrent detach, bail */
> +	if (cur_ioas != access->ioas_unpin)
> +		return -EBUSY;
> +
> +	if (IS_ERR(new_ioas))
> +		return PTR_ERR(new_ioas);
> +
> +	if (cur_ioas == new_ioas) {
> +		/* Do not forget to put since we allow a duplication */
> +		iommufd_put_object(&new_ioas->obj);
> +		return 0;
> +	}
> +
> +	/*
> +	 * Set ioas to NULL to block any further iommufd_access_pin_pages().
> +	 * iommufd_access_unpin_pages() can continue using access->ioas_unpin.
> +	 */
> +	access->ioas = NULL;
> +
> +	if (new_ioas) {
> +		rc = iopt_add_access(&new_ioas->iopt, access);
> +		if (rc) {
> +			iommufd_put_object(&new_ioas->obj);
> +			access->ioas = cur_ioas;
> +			return rc;
> +		}
> +		iommufd_ref_to_users(&new_ioas->obj);

Kevin's suggestion to just open code the refcount_inc here

And have a wrapper func that does:

iommufd_access_change_ioas_id(struct iommufd_access *access, u32 id)
{
	struct iommufd_ioas *ioas = iommufd_get_ioas(ictx, ioas_id);
	int rc;

	if (IS_ERR(ioas))
		return PTR_ERR(ioas);
	rc = iommufd_access_change_ioas(access, ioas);
	iommufd_put_object(&ioas->obj);
	return rc;
}

Does looks cleaner

Then we delete iommufd_ref_to_users() as there are no users (once all
the branches are merged).

Logic looks OK otherwise

Jason