Received-SPF: pass (google.com: domain of linux-kernel+bounces-193467-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
From: Lu Baolu <baolu.lu@linux.intel.com>
To: Joerg Roedel <joro@8bytes.org>,
	Will Deacon <will@kernel.org>,
	Robin Murphy <robin.murphy@arm.com>,
	Jason Gunthorpe <jgg@ziepe.ca>,
	Kevin Tian <kevin.tian@intel.com>
Cc: Yi Liu <yi.l.liu@intel.com>,
	David Airlie <airlied@gmail.com>,
	Daniel Vetter <daniel@ffwll.ch>,
	Kalle Valo <kvalo@kernel.org>,
	Bjorn Andersson <andersson@kernel.org>,
	Mathieu Poirier <mathieu.poirier@linaro.org>,
	Alex Williamson <alex.williamson@redhat.com>,
	mst@redhat.com,
	Jason Wang <jasowang@redhat.com>,
	Thierry Reding <thierry.reding@gmail.com>,
	Jonathan Hunter <jonathanh@nvidia.com>,
	Mikko Perttunen <mperttunen@nvidia.com>,
	iommu@lists.linux.dev,
	dri-devel@lists.freedesktop.org,
	linux-kernel@vger.kernel.org,
	Lu Baolu <baolu.lu@linux.intel.com>
Subject: [PATCH 17/20] iommu/vt-d: Simplify compatibility check for identity domain
Date: Wed, 29 May 2024 13:32:47 +0800
Message-Id: <20240529053250.91284-18-baolu.lu@linux.intel.com>
In-Reply-To: <20240529053250.91284-1-baolu.lu@linux.intel.com>
References: <20240529053250.91284-1-baolu.lu@linux.intel.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

Currently, the identity domain attachment follows the same path as the
paging domain attachment and is subject to the same compatibility checks
as a normal paging domain. However, this level of check is unnecessary
for the identity domain since it only requires the hardware to support
passthrough mode, which is a given for modern hardware.

On the early VT-d platforms, where hardware passthrough mode is not yet
supported, the identity domain is supported by a makeshift paging domain
with the entire system memory 1:1 mapped. For such early hardware, the
appropriate domain type should be returned in device_def_domain_type(),
and the identity domain should be simplified in compatibility checks.

The identity domain workaround in prepare_domain_attach_device() is just
temporary and should be removed once the identity domain is converted to
have its own dedicated attachment path.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/intel/iommu.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
index eb8e08699b80..693a6d7c79ed 100644
--- a/drivers/iommu/intel/iommu.c
+++ b/drivers/iommu/intel/iommu.c
@@ -2190,6 +2190,16 @@ static bool device_rmrr_is_relaxable(struct device *dev)
  */
 static int device_def_domain_type(struct device *dev)
 {
+	struct device_domain_info *info = dev_iommu_priv_get(dev);
+	struct intel_iommu *iommu = info->iommu;
+
+	/*
+	 * Hardware does not support the passthrough translation mode.
+	 * Always use a dynamaic mapping domain.
+	 */
+	if (!ecap_pass_through(iommu->ecap))
+		return IOMMU_DOMAIN_DMA;
+
 	if (dev_is_pci(dev)) {
 		struct pci_dev *pdev = to_pci_dev(dev);
 
@@ -3802,6 +3812,14 @@ int prepare_domain_attach_device(struct iommu_domain *domain,
 	struct intel_iommu *iommu = info->iommu;
 	int addr_width;
 
+	/*
+	 * This is a temporary solution as the identity domain attachment
+	 * goes through this path as well. It should be removed once the
+	 * identity domain has its own attach path.
+	 */
+	if (domain->type == IOMMU_DOMAIN_IDENTITY)
+		return ecap_pass_through(iommu->ecap) ? 0 : -EOPNOTSUPP;
+
 	if (dmar_domain->force_snooping && !ecap_sc_support(iommu->ecap))
 		return -EINVAL;
 
-- 
2.34.1