[PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[Jason Gunthorpe]

iommu_dma_iova_link_swiotlb() processes a mapping that is unaligned in three
parts, the head, middle and trailer. If the middle is empty because there
are no aligned pages it will call down to iommu_map() with a 0 size
which the iommupt implementation will fail as illegal.

It then tries to do an error unwind and starts from the wrong spot
corrupting the mapping so the eventual destruction triggers a WARN_ON.

Check for 0 length and avoid mapping and use offset not 0 as the starting
point to unlink.

This is frequently triggered by using some kinds of thunderbolt NVMe
drives that trigger forced SWIOTLB for unaligned memory. NVMe seems to
pass in oddly aligned buffers for the passthrough commands from smartctl
that hit this condition.

Cc: stable@vger.kernel.org
Fixes: 433a76207dcf ("dma-mapping: Implement link/unlink ranges API")
Reported-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/iommu/dma-iommu.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

This was discovered because iommupt errors on mapping length=0 instead of
making it a NOP, so it is an became an issue since commit d6c65b0fd621
("iommupt: Avoid rewalking during map") making it a regression this merge
window.

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 54d96e847f161b..381b60d9e7ceaf 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -1918,12 +1918,18 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
 			return 0;
 	}
 
+	/*
+	 * After removing the partial head and tail, there may be no aligned
+	 * middle left to map.  The tail still gets bounced below.
+	 */
 	size -= iova_end_pad;
-	error = __dma_iova_link(dev, addr + mapped, phys + mapped, size, dir,
-			attrs);
-	if (error)
-		goto out_unmap;
-	mapped += size;
+	if (size) {
+		error = __dma_iova_link(dev, addr + mapped, phys + mapped,
+				size, dir, attrs);
+		if (error)
+			goto out_unmap;
+		mapped += size;
+	}
 
 	if (iova_end_pad) {
 		error = iommu_dma_iova_bounce_and_link(dev, addr + mapped,
@@ -1936,7 +1942,8 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
 	return 0;
 
 out_unmap:
-	dma_iova_unlink(dev, state, 0, mapped, dir, attrs);
+	if (mapped)
+		dma_iova_unlink(dev, state, offset, mapped, dir, attrs);
 	return error;
 }
 

base-commit: 4549871118cf616eecdd2d939f78e3b9e1dddc48
-- 
2.43.0

Re: [PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[Leon Romanovsky]

On Mon, Jun 08, 2026 at 03:10:04PM -0300, Jason Gunthorpe wrote:
> iommu_dma_iova_link_swiotlb() processes a mapping that is unaligned in three
> parts, the head, middle and trailer. If the middle is empty because there
> are no aligned pages it will call down to iommu_map() with a 0 size
> which the iommupt implementation will fail as illegal.
> 
> It then tries to do an error unwind and starts from the wrong spot
> corrupting the mapping so the eventual destruction triggers a WARN_ON.
> 
> Check for 0 length and avoid mapping and use offset not 0 as the starting
> point to unlink.
> 
> This is frequently triggered by using some kinds of thunderbolt NVMe
> drives that trigger forced SWIOTLB for unaligned memory. NVMe seems to
> pass in oddly aligned buffers for the passthrough commands from smartctl
> that hit this condition.
> 
> Cc: stable@vger.kernel.org
> Fixes: 433a76207dcf ("dma-mapping: Implement link/unlink ranges API")
> Reported-by: Mark Lord <mlord@pobox.com>
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  drivers/iommu/dma-iommu.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
> 
> This was discovered because iommupt errors on mapping length=0 instead of
> making it a NOP, so it is an became an issue since commit d6c65b0fd621
> ("iommupt: Avoid rewalking during map") making it a regression this merge
> window.
> 

Thanks,
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>

Re: [PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[Samiullah Khawaja]

On Mon, Jun 08, 2026 at 03:10:04PM -0300, Jason Gunthorpe wrote:
>iommu_dma_iova_link_swiotlb() processes a mapping that is unaligned in three
>parts, the head, middle and trailer. If the middle is empty because there
>are no aligned pages it will call down to iommu_map() with a 0 size
>which the iommupt implementation will fail as illegal.
>
>It then tries to do an error unwind and starts from the wrong spot
>corrupting the mapping so the eventual destruction triggers a WARN_ON.
>
>Check for 0 length and avoid mapping and use offset not 0 as the starting
>point to unlink.
>
>This is frequently triggered by using some kinds of thunderbolt NVMe
>drives that trigger forced SWIOTLB for unaligned memory. NVMe seems to
>pass in oddly aligned buffers for the passthrough commands from smartctl
>that hit this condition.
>
>Cc: stable@vger.kernel.org
>Fixes: 433a76207dcf ("dma-mapping: Implement link/unlink ranges API")
>Reported-by: Mark Lord <mlord@pobox.com>
>Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
>---
> drivers/iommu/dma-iommu.c | 19 +++++++++++++------
> 1 file changed, 13 insertions(+), 6 deletions(-)
>
>This was discovered because iommupt errors on mapping length=0 instead of
>making it a NOP, so it is an became an issue since commit d6c65b0fd621
>("iommupt: Avoid rewalking during map") making it a regression this merge
>window.
>
>diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
>index 54d96e847f161b..381b60d9e7ceaf 100644
>--- a/drivers/iommu/dma-iommu.c
>+++ b/drivers/iommu/dma-iommu.c
>@@ -1918,12 +1918,18 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
> 			return 0;
> 	}
>
>+	/*
>+	 * After removing the partial head and tail, there may be no aligned
>+	 * middle left to map.  The tail still gets bounced below.
>+	 */
> 	size -= iova_end_pad;
>-	error = __dma_iova_link(dev, addr + mapped, phys + mapped, size, dir,
>-			attrs);
>-	if (error)
>-		goto out_unmap;
>-	mapped += size;
>+	if (size) {
>+		error = __dma_iova_link(dev, addr + mapped, phys + mapped,
>+				size, dir, attrs);
>+		if (error)
>+			goto out_unmap;
>+		mapped += size;
>+	}
>
> 	if (iova_end_pad) {
> 		error = iommu_dma_iova_bounce_and_link(dev, addr + mapped,
>@@ -1936,7 +1942,8 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
> 	return 0;
>
> out_unmap:
>-	dma_iova_unlink(dev, state, 0, mapped, dir, attrs);
>+	if (mapped)
>+		dma_iova_unlink(dev, state, offset, mapped, dir, attrs);
> 	return error;
> }
>
>
>base-commit: 4549871118cf616eecdd2d939f78e3b9e1dddc48
>-- 
>2.43.0
>
>

Reviewed-by: Samiullah Khawaja <skhawaja@google.com>

Re: [PATCH rc] iommu/dma: Do not try to iommu_map a 0 length region in swiotlb

[Marek Szyprowski]

On 08.06.2026 20:10, Jason Gunthorpe wrote:
> iommu_dma_iova_link_swiotlb() processes a mapping that is unaligned in three
> parts, the head, middle and trailer. If the middle is empty because there
> are no aligned pages it will call down to iommu_map() with a 0 size
> which the iommupt implementation will fail as illegal.
>
> It then tries to do an error unwind and starts from the wrong spot
> corrupting the mapping so the eventual destruction triggers a WARN_ON.
>
> Check for 0 length and avoid mapping and use offset not 0 as the starting
> point to unlink.
>
> This is frequently triggered by using some kinds of thunderbolt NVMe
> drives that trigger forced SWIOTLB for unaligned memory. NVMe seems to
> pass in oddly aligned buffers for the passthrough commands from smartctl
> that hit this condition.
>
> Cc: stable@vger.kernel.org
> Fixes: 433a76207dcf ("dma-mapping: Implement link/unlink ranges API")
> Reported-by: Mark Lord <mlord@pobox.com>
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Applied to dma-mapping-fixes, thanks!


> ---
>  drivers/iommu/dma-iommu.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
>
> This was discovered because iommupt errors on mapping length=0 instead of
> making it a NOP, so it is an became an issue since commit d6c65b0fd621
> ("iommupt: Avoid rewalking during map") making it a regression this merge
> window.
>
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 54d96e847f161b..381b60d9e7ceaf 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -1918,12 +1918,18 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
>  			return 0;
>  	}
>  
> +	/*
> +	 * After removing the partial head and tail, there may be no aligned
> +	 * middle left to map.  The tail still gets bounced below.
> +	 */
>  	size -= iova_end_pad;
> -	error = __dma_iova_link(dev, addr + mapped, phys + mapped, size, dir,
> -			attrs);
> -	if (error)
> -		goto out_unmap;
> -	mapped += size;
> +	if (size) {
> +		error = __dma_iova_link(dev, addr + mapped, phys + mapped,
> +				size, dir, attrs);
> +		if (error)
> +			goto out_unmap;
> +		mapped += size;
> +	}
>  
>  	if (iova_end_pad) {
>  		error = iommu_dma_iova_bounce_and_link(dev, addr + mapped,
> @@ -1936,7 +1942,8 @@ static int iommu_dma_iova_link_swiotlb(struct device *dev,
>  	return 0;
>  
>  out_unmap:
> -	dma_iova_unlink(dev, state, 0, mapped, dir, attrs);
> +	if (mapped)
> +		dma_iova_unlink(dev, state, offset, mapped, dir, attrs);
>  	return error;
>  }
>  
>
> base-commit: 4549871118cf616eecdd2d939f78e3b9e1dddc48

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland