Received-SPF: pass (google.com: domain of linux-kernel+bounces-111832-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Fri, 22 Mar 2024 13:59:27 -0300
From: Jason Gunthorpe <jgg@ziepe.ca>
To: Baolu Lu <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: <20240322165927.GG66976@ziepe.ca>
References: <20240122073903.24406-1-baolu.lu@linux.intel.com>
 <20240122073903.24406-3-baolu.lu@linux.intel.com>
 <20240308174605.GV9225@ziepe.ca>
 <b33bf29b-2fe5-4a2a-a2ce-9fd8d67c5f6f@linux.intel.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b33bf29b-2fe5-4a2a-a2ce-9fd8d67c5f6f@linux.intel.com>

On Thu, Mar 14, 2024 at 03:41:23PM +0800, Baolu Lu wrote:

> The whole cookie mechanism aims to address two things:
> 
> - Extend the domain lifetime until all pending page faults are
> resolved.

Like you answered, I think the flush is a simpler scheme..

> - Associate information about the iommu device with each attachment of
>   the domain so that the iommufd device object ID could be quickly
>   retrieved in the fault delivering path.
 
> > 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..
> 
> It's not getting back hwpt from domain, just getting the iommufd_device
> in the fault delivering path. The iommufd_device is not per-domain, but
> per-domain-attachment.

It does make sense you'd need that, but I think something like this is
a more direct way to get it. Caller allocates the handle struct. The
iopf will provide the handle from the XA to the
callback. container_of() not void * is used to in the caller's API.

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 68e648b5576706..0d29d8f0847cd9 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -46,6 +46,8 @@ static unsigned int iommu_def_domain_type __read_mostly;
 static bool iommu_dma_strict __read_mostly = IS_ENABLED(CONFIG_IOMMU_DEFAULT_DMA_STRICT);
 static u32 iommu_cmd_line __read_mostly;
 
+enum { IOMMU_PASID_ARRAY_DOMAIN, IOMMU_PASID_ARRAY_HANDLE };
+
 struct iommu_group {
 	struct kobject kobj;
 	struct kobject *devices_kobj;
@@ -3516,18 +3518,20 @@ static void __iommu_remove_group_pasid(struct iommu_group *group,
 }
 
 /*
- * iommu_attach_device_pasid() - Attach a domain to pasid of device
+ * __iommu_attach_device_pasid() - Attach a domain to pasid of device
  * @domain: the iommu domain.
  * @dev: the attached device.
  * @pasid: the pasid of the device.
  *
  * Return: 0 on success, or an error.
  */
-int iommu_attach_device_pasid(struct iommu_domain *domain,
-			      struct device *dev, ioasid_t pasid)
+int __iommu_attach_device_pasid(struct iommu_domain *domain, struct device *dev,
+				ioasid_t pasid,
+				struct iommu_pasid_handle *handle)
 {
 	/* Caller must be a probed driver on dev */
 	struct iommu_group *group = dev->iommu_group;
+	void *xa_entry;
 	void *curr;
 	int ret;
 
@@ -3541,9 +3545,14 @@ int iommu_attach_device_pasid(struct iommu_domain *domain,
 		return -EINVAL;
 
 	mutex_lock(&group->mutex);
-	curr = xa_cmpxchg(&group->pasid_array, pasid, NULL, domain, GFP_KERNEL);
+	if (handle)
+		xa_entry = xa_tag_pointer(handle, IOMMU_PASID_ARRAY_HANDLE);
+	else
+		xa_entry = xa_tag_pointer(domain, IOMMU_PASID_ARRAY_DOMAIN);
+	curr = xa_cmpxchg(&group->pasid_array, pasid, NULL, xa_entry,
+			  GFP_KERNEL);
 	if (curr) {
-		ret = xa_err(curr) ? : -EBUSY;
+		ret = xa_err(curr) ?: -EBUSY;
 		goto out_unlock;
 	}
 
@@ -3556,7 +3565,7 @@ int iommu_attach_device_pasid(struct iommu_domain *domain,
 	mutex_unlock(&group->mutex);
 	return ret;
 }
-EXPORT_SYMBOL_GPL(iommu_attach_device_pasid);
+EXPORT_SYMBOL_GPL(__iommu_attach_device_pasid);
 
 /*
  * iommu_detach_device_pasid() - Detach the domain from pasid of device
@@ -3600,13 +3609,23 @@ struct iommu_domain *iommu_get_domain_for_dev_pasid(struct device *dev,
 {
 	/* Caller must be a probed driver on dev */
 	struct iommu_group *group = dev->iommu_group;
+	struct iommu_pasid_handle *handle;
 	struct iommu_domain *domain;
+	void *xa_entry;
 
 	if (!group)
 		return NULL;
 
 	xa_lock(&group->pasid_array);
-	domain = xa_load(&group->pasid_array, pasid);
+	xa_entry = xa_load(&group->pasid_array, pasid);
+	if (xa_pointer_tag(xa_entry) == IOMMU_PASID_ARRAY_HANDLE) {
+		handle = xa_untag_pointer(xa_entry);
+		domain = handle->domain;
+	} else if (xa_pointer_tag(xa_entry) == IOMMU_PASID_ARRAY_DOMAIN) {
+		domain = xa_untag_pointer(xa_entry);
+	} else {
+		domain = NULL;
+	}
 	if (type && domain && domain->type != type)
 		domain = ERR_PTR(-EBUSY);
 	xa_unlock(&group->pasid_array);
diff --git a/include/linux/iommu.h b/include/linux/iommu.h
index 1ea2a820e1eb03..47c121d4e13f65 100644
--- a/include/linux/iommu.h
+++ b/include/linux/iommu.h
@@ -947,8 +947,27 @@ void iommu_device_release_dma_owner(struct device *dev);
 
 struct iommu_domain *iommu_sva_domain_alloc(struct device *dev,
 					    struct mm_struct *mm);
-int iommu_attach_device_pasid(struct iommu_domain *domain,
-			      struct device *dev, ioasid_t pasid);
+
+struct iommu_pasid_handle {
+	struct iommu_domain *domain;
+};
+
+int __iommu_attach_device_pasid(struct iommu_domain *domain, struct device *dev,
+				ioasid_t pasid,
+				struct iommu_pasid_handle *handle);
+
+static inline int iommu_attach_device_pasid(struct iommu_domain *domain,
+					    struct device *dev, ioasid_t pasid)
+{
+	return __iommu_attach_device_pasid(domain, dev, pasid, NULL);
+}
+static inline int
+iommu_attach_device_pasid_handle(struct iommu_pasid_handle *handle,
+				 struct device *dev, ioasid_t pasid)
+{
+	return __iommu_attach_device_pasid(handle->domain, dev, pasid, handle);
+}
+
 void iommu_detach_device_pasid(struct iommu_domain *domain,
 			       struct device *dev, ioasid_t pasid);
 struct iommu_domain *