Received-SPF: pass (google.com: domain of linux-kernel+bounces-64386-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates
 216.228.117.160 as permitted sender) receiver=protection.outlook.com;
 client-ip=216.228.117.160; helo=mail.nvidia.com; pr=C
From: Nicolin Chen <nicolinc@nvidia.com>
To: <sagi@grimberg.me>, <hch@lst.de>, <axboe@kernel.dk>, <kbusch@kernel.org>,
	<will@kernel.org>, <joro@8bytes.org>, <robin.murphy@arm.com>
CC: <jgg@nvidia.com>, <linux-nvme@lists.infradead.org>,
	<linux-kernel@vger.kernel.org>, <iommu@lists.linux.dev>, <murphyt7@tcd.ie>,
	<baolu.lu@linux.intel.com>
Subject: [PATCH v1 2/2] nvme-pci: Fix iommu map (via swiotlb) failures when PAGE_SIZE=64KB
Date: Tue, 13 Feb 2024 13:53:57 -0800
Message-ID: <60bdcc29a2bcf12c6ab95cf0ea480d67c41c51e7.1707851466.git.nicolinc@nvidia.com>
In-Reply-To: <cover.1707851466.git.nicolinc@nvidia.com>
References: <cover.1707851466.git.nicolinc@nvidia.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 13 Feb 2024 21:54:15.3834
 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: fdd58479-14a4-463c-9e79-08dc2cde525b
X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a;Ip=[216.228.117.160];Helo=[mail.nvidia.com]
X-MS-Exchange-CrossTenant-AuthSource:
	BN2PEPF000044AA.namprd04.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: DM4PR12MB7576

The min_align_mask is set for swiotlb_max_mapping_size calculation used
by dma_opt_mapping_size in the a few lines behind.

By default, the swiotlb_max_mapping_size() returns 256KB subtracted by
roundup(min_align_mask, IO_TLB_SIZE). Since NVME_CTRL_PAGE_SIZE=4KB, in
this case the swiotlb_max_mapping_size() returns 252KB (256KB - 4KB).

Meanwhile, the alloc_size eventually passed in to swiotlb_tbl_map_single
in dma-iommu is aligned with its attaching domain's iova->granule. If the
domain (backed by an IO page table) is using a 64KB page size, alloc_size
can still ask for 256KB v.s. 252KB, which fails the mapping request:
    systemd[1]: Started Journal Service.
 => nvme 0000:00:01.0: swiotlb buffer is full (sz: 262144 bytes), total 32768 (slots), used 128 (slots)
    note: journal-offline[392] exited with irqs disabled
    note: journal-offline[392] exited with preempt_count 1

Another factor is that the attached domain of an NVME device can change,
so can the value of iova->granule, meaning that the min_align_mask cannot
be set simply using the iova->granule value from its attached domain.

Since the iova->granule usually picks the smallest one from pgsize_bitmap
and it oftens matches with CPU's PAGE_SIZE, simply set min_align_mask to
"PAGE_SIZE - 1".

Note that the other patch "iommu/dma: Force swiotlb_max_mapping_size on
an untrusted device" is required to entirely fix this mapping failure.

Fixes: 82612d66d51d ("iommu: Allow the dma-iommu api to use bounce buffers")
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
---
 drivers/nvme/host/pci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index e6267a6aa380..ad41fe0bf2e3 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -2967,7 +2967,7 @@ static struct nvme_dev *nvme_pci_alloc_dev(struct pci_dev *pdev,
 		dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48));
 	else
 		dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
-	dma_set_min_align_mask(&pdev->dev, NVME_CTRL_PAGE_SIZE - 1);
+	dma_set_min_align_mask(&pdev->dev, PAGE_SIZE - 1);
 	dma_set_max_seg_size(&pdev->dev, 0xffffffff);
 
 	/*
-- 
2.43.0