Received-SPF: pass (google.com: domain of linux-kernel+bounces-97412-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Fri, 8 Mar 2024 13:46:05 -0400
From: Jason Gunthorpe <jgg@ziepe.ca>
To: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>, Joerg Roedel <joro@8bytes.org>,
	Will Deacon <will@kernel.org>, Robin Murphy <robin.murphy@arm.com>,
	Jean-Philippe Brucker <jean-philippe@linaro.org>,
	Nicolin Chen <nicolinc@nvidia.com>, Yi Liu <yi.l.liu@intel.com>,
	Jacob Pan <jacob.jun.pan@linux.intel.com>,
	Joel Granados <j.granados@samsung.com>, iommu@lists.linux.dev,
	virtualization@lists.linux-foundation.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3 2/8] iommu/sva: Use iopf domain attach/detach interface
Message-ID: <20240308174605.GV9225@ziepe.ca>
References: <20240122073903.24406-1-baolu.lu@linux.intel.com>
 <20240122073903.24406-3-baolu.lu@linux.intel.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240122073903.24406-3-baolu.lu@linux.intel.com>

On Mon, Jan 22, 2024 at 03:38:57PM +0800, Lu Baolu wrote:
> @@ -215,7 +202,23 @@ static struct iopf_group *iopf_group_alloc(struct iommu_fault_param *iopf_param,
>  		group = abort_group;
>  	}
>  
> +	cookie = iopf_pasid_cookie_get(iopf_param->dev, pasid);
> +	if (!cookie && pasid != IOMMU_NO_PASID)
> +		cookie = iopf_pasid_cookie_get(iopf_param->dev, IOMMU_NO_PASID);
> +	if (IS_ERR(cookie) || !cookie) {
> +		/*
> +		 * The PASID of this device was not attached by an I/O-capable
> +		 * domain. Ask the caller to abort handling of this fault.
> +		 * Otherwise, the reference count will be switched to the new
> +		 * iopf group and will be released in iopf_free_group().
> +		 */
> +		kfree(group);
> +		group = abort_group;
> +		cookie = NULL;
> +	}

If this is the main point of the cookie mechansim - why not just have
a refcount inside the domain itself?

I'm really having a hard time making sense of this cookie thing, we
have a lifetime issue on the domain pointer, why is adding another
structure the answer?

I see we also need to stick a pointer in the domain for iommufd to get
back to the hwpt, but that doesn't seem to need such a big system to
accomplish - just add a void *. It would make sense for the domain to
have some optional pointer to a struct for all the fault related data
that becomes allocated when a PRI domain is created..

Also, I thought we'd basically said that domain detatch is supposed to
flush any open PRIs before returning, what happened to that as a
solution to the lifetime problem?

Regardless I think we should split this into two series - improve the PRI
lifetime model for domains and then put iommufd on top of it..

Jason