1. Patchew
  2. linux
  3. [PATCH v2 0/4] dma-buf: document revoke mechanism to invalidate shared buffers
  4. View patch

[PATCH v2 3/4] iommufd: Require DMABUF revoke semantics

Leon Romanovsky posted 4 patches 3 weeks, 1 day ago
[PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Leon Romanovsky 3 weeks, 1 day ago 
From: Leon Romanovsky <leonro@nvidia.com>

IOMMUFD does not support page fault handling, and after a call to
.invalidate_mappings() all mappings become invalid. Ensure that
the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
(for example, VFIO).

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/iommu/iommufd/pages.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
index 76f900fa1687..a5eb2bc4ef48 100644
--- a/drivers/iommu/iommufd/pages.c
+++ b/drivers/iommu/iommufd/pages.c
@@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
 		mutex_unlock(&pages->mutex);
 	}
 
-	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
+	rc = dma_buf_pin(attach);
 	if (rc)
 		goto err_detach;
 
+	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
+	if (rc)
+		goto err_unpin;
+
 	dma_resv_unlock(dmabuf->resv);
 
 	/* On success iopt_release_pages() will detach and put the dmabuf. */
 	pages->dmabuf.attach = attach;
 	return 0;
 
+err_unpin:
+	dma_buf_unpin(attach);
 err_detach:
 	dma_resv_unlock(dmabuf->resv);
 	dma_buf_detach(dmabuf, attach);
@@ -1656,6 +1662,7 @@ void iopt_release_pages(struct kref *kref)
 	if (iopt_is_dmabuf(pages) && pages->dmabuf.attach) {
 		struct dma_buf *dmabuf = pages->dmabuf.attach->dmabuf;
 
+		dma_buf_unpin(pages->dmabuf.attach);
 		dma_buf_detach(dmabuf, pages->dmabuf.attach);
 		dma_buf_put(dmabuf);
 		WARN_ON(!list_empty(&pages->dmabuf.tracker));

-- 
2.52.0
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Jason Gunthorpe 2 weeks, 6 days ago 
On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> IOMMUFD does not support page fault handling, and after a call to
> .invalidate_mappings() all mappings become invalid. Ensure that
> the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> (for example, VFIO).
> 
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>  drivers/iommu/iommufd/pages.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> index 76f900fa1687..a5eb2bc4ef48 100644
> --- a/drivers/iommu/iommufd/pages.c
> +++ b/drivers/iommu/iommufd/pages.c
> @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
>  		mutex_unlock(&pages->mutex);
>  	}
>  
> -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> +	rc = dma_buf_pin(attach);
>  	if (rc)
>  		goto err_detach;
>  
> +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> +	if (rc)
> +		goto err_unpin;
> +
>  	dma_resv_unlock(dmabuf->resv);
>  
>  	/* On success iopt_release_pages() will detach and put the dmabuf. */
>  	pages->dmabuf.attach = attach;
>  	return 0;

Don't we need an explicit unpin after unmapping?

Jason
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Leon Romanovsky 2 weeks, 6 days ago 
On Mon, Jan 19, 2026 at 12:59:51PM -0400, Jason Gunthorpe wrote:
> On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro@nvidia.com>
> > 
> > IOMMUFD does not support page fault handling, and after a call to
> > .invalidate_mappings() all mappings become invalid. Ensure that
> > the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> > (for example, VFIO).
> > 
> > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > ---
> >  drivers/iommu/iommufd/pages.c | 9 ++++++++-
> >  1 file changed, 8 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> > index 76f900fa1687..a5eb2bc4ef48 100644
> > --- a/drivers/iommu/iommufd/pages.c
> > +++ b/drivers/iommu/iommufd/pages.c
> > @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
> >  		mutex_unlock(&pages->mutex);
> >  	}
> >  
> > -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > +	rc = dma_buf_pin(attach);
> >  	if (rc)
> >  		goto err_detach;
> >  
> > +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > +	if (rc)
> > +		goto err_unpin;
> > +
> >  	dma_resv_unlock(dmabuf->resv);
> >  
> >  	/* On success iopt_release_pages() will detach and put the dmabuf. */
> >  	pages->dmabuf.attach = attach;
> >  	return 0;
> 
> Don't we need an explicit unpin after unmapping?

Yes, but this patch is going to be dropped in v3 because of this
suggestion.
https://lore.kernel.org/all/a397ff1e-615f-4873-98a9-940f9c16f85c@amd.com

Thanks


> 
> Jason
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Jason Gunthorpe 2 weeks, 6 days ago 
On Mon, Jan 19, 2026 at 08:23:00PM +0200, Leon Romanovsky wrote:
> On Mon, Jan 19, 2026 at 12:59:51PM -0400, Jason Gunthorpe wrote:
> > On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> > > From: Leon Romanovsky <leonro@nvidia.com>
> > > 
> > > IOMMUFD does not support page fault handling, and after a call to
> > > .invalidate_mappings() all mappings become invalid. Ensure that
> > > the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> > > (for example, VFIO).
> > > 
> > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > > ---
> > >  drivers/iommu/iommufd/pages.c | 9 ++++++++-
> > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> > > index 76f900fa1687..a5eb2bc4ef48 100644
> > > --- a/drivers/iommu/iommufd/pages.c
> > > +++ b/drivers/iommu/iommufd/pages.c
> > > @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
> > >  		mutex_unlock(&pages->mutex);
> > >  	}
> > >  
> > > -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > +	rc = dma_buf_pin(attach);
> > >  	if (rc)
> > >  		goto err_detach;
> > >  
> > > +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > +	if (rc)
> > > +		goto err_unpin;
> > > +
> > >  	dma_resv_unlock(dmabuf->resv);
> > >  
> > >  	/* On success iopt_release_pages() will detach and put the dmabuf. */
> > >  	pages->dmabuf.attach = attach;
> > >  	return 0;
> > 
> > Don't we need an explicit unpin after unmapping?
> 
> Yes, but this patch is going to be dropped in v3 because of this
> suggestion.
> https://lore.kernel.org/all/a397ff1e-615f-4873-98a9-940f9c16f85c@amd.com

That's not right, that suggestion is about changing VFIO. iommufd must
still act as a pinning importer!

Jason
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Leon Romanovsky 2 weeks, 5 days ago 
On Mon, Jan 19, 2026 at 03:54:44PM -0400, Jason Gunthorpe wrote:
> On Mon, Jan 19, 2026 at 08:23:00PM +0200, Leon Romanovsky wrote:
> > On Mon, Jan 19, 2026 at 12:59:51PM -0400, Jason Gunthorpe wrote:
> > > On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> > > > From: Leon Romanovsky <leonro@nvidia.com>
> > > > 
> > > > IOMMUFD does not support page fault handling, and after a call to
> > > > .invalidate_mappings() all mappings become invalid. Ensure that
> > > > the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> > > > (for example, VFIO).
> > > > 
> > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > > > ---
> > > >  drivers/iommu/iommufd/pages.c | 9 ++++++++-
> > > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> > > > index 76f900fa1687..a5eb2bc4ef48 100644
> > > > --- a/drivers/iommu/iommufd/pages.c
> > > > +++ b/drivers/iommu/iommufd/pages.c
> > > > @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
> > > >  		mutex_unlock(&pages->mutex);
> > > >  	}
> > > >  
> > > > -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > +	rc = dma_buf_pin(attach);
> > > >  	if (rc)
> > > >  		goto err_detach;
> > > >  
> > > > +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > +	if (rc)
> > > > +		goto err_unpin;
> > > > +
> > > >  	dma_resv_unlock(dmabuf->resv);
> > > >  
> > > >  	/* On success iopt_release_pages() will detach and put the dmabuf. */
> > > >  	pages->dmabuf.attach = attach;
> > > >  	return 0;
> > > 
> > > Don't we need an explicit unpin after unmapping?
> > 
> > Yes, but this patch is going to be dropped in v3 because of this
> > suggestion.
> > https://lore.kernel.org/all/a397ff1e-615f-4873-98a9-940f9c16f85c@amd.com
> 
> That's not right, that suggestion is about changing VFIO. iommufd must
> still act as a pinning importer!

There is no change in iommufd, as it invokes dma_buf_dynamic_attach()
with a valid &iopt_dmabuf_attach_revoke_ops. The check determining whether
iommufd can perform a revoke is handled there.

Thanks

> 
> Jason
>
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Jason Gunthorpe 2 weeks, 5 days ago 
On Tue, Jan 20, 2026 at 03:10:46PM +0200, Leon Romanovsky wrote:
> On Mon, Jan 19, 2026 at 03:54:44PM -0400, Jason Gunthorpe wrote:
> > On Mon, Jan 19, 2026 at 08:23:00PM +0200, Leon Romanovsky wrote:
> > > On Mon, Jan 19, 2026 at 12:59:51PM -0400, Jason Gunthorpe wrote:
> > > > On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> > > > > From: Leon Romanovsky <leonro@nvidia.com>
> > > > > 
> > > > > IOMMUFD does not support page fault handling, and after a call to
> > > > > .invalidate_mappings() all mappings become invalid. Ensure that
> > > > > the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> > > > > (for example, VFIO).
> > > > > 
> > > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > > > > ---
> > > > >  drivers/iommu/iommufd/pages.c | 9 ++++++++-
> > > > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> > > > > index 76f900fa1687..a5eb2bc4ef48 100644
> > > > > --- a/drivers/iommu/iommufd/pages.c
> > > > > +++ b/drivers/iommu/iommufd/pages.c
> > > > > @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
> > > > >  		mutex_unlock(&pages->mutex);
> > > > >  	}
> > > > >  
> > > > > -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > > +	rc = dma_buf_pin(attach);
> > > > >  	if (rc)
> > > > >  		goto err_detach;
> > > > >  
> > > > > +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > > +	if (rc)
> > > > > +		goto err_unpin;
> > > > > +
> > > > >  	dma_resv_unlock(dmabuf->resv);
> > > > >  
> > > > >  	/* On success iopt_release_pages() will detach and put the dmabuf. */
> > > > >  	pages->dmabuf.attach = attach;
> > > > >  	return 0;
> > > > 
> > > > Don't we need an explicit unpin after unmapping?
> > > 
> > > Yes, but this patch is going to be dropped in v3 because of this
> > > suggestion.
> > > https://lore.kernel.org/all/a397ff1e-615f-4873-98a9-940f9c16f85c@amd.com
> > 
> > That's not right, that suggestion is about changing VFIO. iommufd must
> > still act as a pinning importer!
> 
> There is no change in iommufd, as it invokes dma_buf_dynamic_attach()
> with a valid &iopt_dmabuf_attach_revoke_ops. The check determining whether
> iommufd can perform a revoke is handled there.

iommufd is a pining importer. I did not add a call to pin because it
only worked with VFIO that would not support it. Now that this series
fixes it the pin must be added. Don't drop this patch.

All the explanations we just gave say this special revoke mode only
activates if the buffer is pinned by the importer, so iommufd must pin
it. Otherwise it says it is working in the move mode with faulting
that it cannot support.

Jason
Re: [PATCH v2 3/4] iommufd: Require DMABUF revoke semantics
Posted by Leon Romanovsky 2 weeks, 5 days ago 
On Tue, Jan 20, 2026 at 09:15:30AM -0400, Jason Gunthorpe wrote:
> On Tue, Jan 20, 2026 at 03:10:46PM +0200, Leon Romanovsky wrote:
> > On Mon, Jan 19, 2026 at 03:54:44PM -0400, Jason Gunthorpe wrote:
> > > On Mon, Jan 19, 2026 at 08:23:00PM +0200, Leon Romanovsky wrote:
> > > > On Mon, Jan 19, 2026 at 12:59:51PM -0400, Jason Gunthorpe wrote:
> > > > > On Sun, Jan 18, 2026 at 02:08:47PM +0200, Leon Romanovsky wrote:
> > > > > > From: Leon Romanovsky <leonro@nvidia.com>
> > > > > > 
> > > > > > IOMMUFD does not support page fault handling, and after a call to
> > > > > > .invalidate_mappings() all mappings become invalid. Ensure that
> > > > > > the IOMMUFD DMABUF importer is bound to a revoke‑aware DMABUF exporter
> > > > > > (for example, VFIO).
> > > > > > 
> > > > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > > > > > ---
> > > > > >  drivers/iommu/iommufd/pages.c | 9 ++++++++-
> > > > > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> > > > > > index 76f900fa1687..a5eb2bc4ef48 100644
> > > > > > --- a/drivers/iommu/iommufd/pages.c
> > > > > > +++ b/drivers/iommu/iommufd/pages.c
> > > > > > @@ -1501,16 +1501,22 @@ static int iopt_map_dmabuf(struct iommufd_ctx *ictx, struct iopt_pages *pages,
> > > > > >  		mutex_unlock(&pages->mutex);
> > > > > >  	}
> > > > > >  
> > > > > > -	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > > > +	rc = dma_buf_pin(attach);
> > > > > >  	if (rc)
> > > > > >  		goto err_detach;
> > > > > >  
> > > > > > +	rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
> > > > > > +	if (rc)
> > > > > > +		goto err_unpin;
> > > > > > +
> > > > > >  	dma_resv_unlock(dmabuf->resv);
> > > > > >  
> > > > > >  	/* On success iopt_release_pages() will detach and put the dmabuf. */
> > > > > >  	pages->dmabuf.attach = attach;
> > > > > >  	return 0;
> > > > > 
> > > > > Don't we need an explicit unpin after unmapping?
> > > > 
> > > > Yes, but this patch is going to be dropped in v3 because of this
> > > > suggestion.
> > > > https://lore.kernel.org/all/a397ff1e-615f-4873-98a9-940f9c16f85c@amd.com
> > > 
> > > That's not right, that suggestion is about changing VFIO. iommufd must
> > > still act as a pinning importer!
> > 
> > There is no change in iommufd, as it invokes dma_buf_dynamic_attach()
> > with a valid &iopt_dmabuf_attach_revoke_ops. The check determining whether
> > iommufd can perform a revoke is handled there.
> 
> iommufd is a pining importer. I did not add a call to pin because it
> only worked with VFIO that would not support it. Now that this series
> fixes it the pin must be added. Don't drop this patch.

No problem, let's keep it.

Thanks

Patchew 2.1-devel-b67e4c2

© 2016 - 2026 Red Hat, Inc.