diff mbox series

[v2,3/3] vfio/pci: Allow MMIO regions to be exported through dma-buf

Message ID 20240624065552.1572580-4-vivek.kasireddy@intel.com (mailing list archive)
State New, archived
Headers show
Series vfio/pci: Allow MMIO regions to be exported through dma-buf | expand

Commit Message

Kasireddy, Vivek June 24, 2024, 6:53 a.m. UTC
From Jason Gunthorpe:
"dma-buf has become a way to safely acquire a handle to non-struct page
memory that can still have lifetime controlled by the exporter. Notably
RDMA can now import dma-buf FDs and build them into MRs which allows for
PCI P2P operations. Extend this to allow vfio-pci to export MMIO memory
from PCI device BARs.

The patch design loosely follows the pattern in commit
db1a8dd916aa ("habanalabs: add support for dma-buf exporter") except this
does not support pinning.

Instead, this implements what, in the past, we've called a revocable
attachment using move. In normal situations the attachment is pinned, as a
BAR does not change physical address. However when the VFIO device is
closed, or a PCI reset is issued, access to the MMIO memory is revoked.

Revoked means that move occurs, but an attempt to immediately re-map the
memory will fail. In the reset case a future move will be triggered when
MMIO access returns. As both close and reset are under userspace control
it is expected that userspace will suspend use of the dma-buf before doing
these operations, the revoke is purely for kernel self-defense against a
hostile userspace."

Following enhancements are made to the original patch:
- Add support for creating dmabuf from multiple areas (or ranges)
- Add a mmap handler to provide CPU access to the dmabuf

Original-patch-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
---
 drivers/vfio/pci/Makefile          |   1 +
 drivers/vfio/pci/dma_buf.c         | 438 +++++++++++++++++++++++++++++
 drivers/vfio/pci/vfio_pci_config.c |  22 +-
 drivers/vfio/pci/vfio_pci_core.c   |  20 +-
 drivers/vfio/pci/vfio_pci_priv.h   |  23 ++
 include/linux/vfio_pci_core.h      |   1 +
 include/uapi/linux/vfio.h          |  25 ++
 7 files changed, 525 insertions(+), 5 deletions(-)
 create mode 100644 drivers/vfio/pci/dma_buf.c

Comments

Daniel Vetter June 24, 2024, 10:02 a.m. UTC | #1
On Sun, Jun 23, 2024 at 11:53:11PM -0700, Vivek Kasireddy wrote:
> From Jason Gunthorpe:
> "dma-buf has become a way to safely acquire a handle to non-struct page
> memory that can still have lifetime controlled by the exporter. Notably
> RDMA can now import dma-buf FDs and build them into MRs which allows for
> PCI P2P operations. Extend this to allow vfio-pci to export MMIO memory
> from PCI device BARs.
> 
> The patch design loosely follows the pattern in commit
> db1a8dd916aa ("habanalabs: add support for dma-buf exporter") except this
> does not support pinning.
> 
> Instead, this implements what, in the past, we've called a revocable
> attachment using move. In normal situations the attachment is pinned, as a
> BAR does not change physical address. However when the VFIO device is
> closed, or a PCI reset is issued, access to the MMIO memory is revoked.
> 
> Revoked means that move occurs, but an attempt to immediately re-map the
> memory will fail. In the reset case a future move will be triggered when
> MMIO access returns. As both close and reset are under userspace control
> it is expected that userspace will suspend use of the dma-buf before doing
> these operations, the revoke is purely for kernel self-defense against a
> hostile userspace."
> 
> Following enhancements are made to the original patch:
> - Add support for creating dmabuf from multiple areas (or ranges)
> - Add a mmap handler to provide CPU access to the dmabuf
> 
> Original-patch-by: Jason Gunthorpe <jgg@nvidia.com>
> Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>

I think this makes sense for dma-buf, so has my Ack. Bunch of things I've
spotted below, but overall looks all fine conceptually. Also a suggestion
for a dma-buf doc patch, so that we can collect acks from at least all
current dynamic importers whether they're code will be fine with this new
dma-buf.

For the actual use-case and upstream merging I think that's up to vfio and
rdma maintainers, at least to me doesn't feel it's impacting the gpu side.
Well maybe long-term people will also use this for gpus, there's
difinitely interest in direct upload to vram, but that's maybe more a
use-case where the nvme side will import gpu buffers.

Cheers, Sima
> ---
>  drivers/vfio/pci/Makefile          |   1 +
>  drivers/vfio/pci/dma_buf.c         | 438 +++++++++++++++++++++++++++++
>  drivers/vfio/pci/vfio_pci_config.c |  22 +-
>  drivers/vfio/pci/vfio_pci_core.c   |  20 +-
>  drivers/vfio/pci/vfio_pci_priv.h   |  23 ++
>  include/linux/vfio_pci_core.h      |   1 +
>  include/uapi/linux/vfio.h          |  25 ++
>  7 files changed, 525 insertions(+), 5 deletions(-)
>  create mode 100644 drivers/vfio/pci/dma_buf.c
> 
> diff --git a/drivers/vfio/pci/Makefile b/drivers/vfio/pci/Makefile
> index cf00c0a7e55c..0cfdc9ede82f 100644
> --- a/drivers/vfio/pci/Makefile
> +++ b/drivers/vfio/pci/Makefile
> @@ -2,6 +2,7 @@
>  
>  vfio-pci-core-y := vfio_pci_core.o vfio_pci_intrs.o vfio_pci_rdwr.o vfio_pci_config.o
>  vfio-pci-core-$(CONFIG_VFIO_PCI_ZDEV_KVM) += vfio_pci_zdev.o
> +vfio-pci-core-$(CONFIG_DMA_SHARED_BUFFER) += dma_buf.o
>  obj-$(CONFIG_VFIO_PCI_CORE) += vfio-pci-core.o
>  
>  vfio-pci-y := vfio_pci.o
> diff --git a/drivers/vfio/pci/dma_buf.c b/drivers/vfio/pci/dma_buf.c
> new file mode 100644
> index 000000000000..68760677c6a0
> --- /dev/null
> +++ b/drivers/vfio/pci/dma_buf.c
> @@ -0,0 +1,438 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/* Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES.
> + */
> +#include <linux/dma-buf.h>
> +#include <linux/pci-p2pdma.h>
> +#include <linux/dma-resv.h>
> +
> +#include "vfio_pci_priv.h"
> +
> +MODULE_IMPORT_NS(DMA_BUF);
> +
> +struct vfio_pci_dma_buf {
> +	struct dma_buf *dmabuf;
> +	struct vfio_pci_core_device *vdev;
> +	struct list_head dmabufs_elm;
> +	unsigned int nr_ranges;
> +	struct vfio_region_dma_range *dma_ranges;
> +	unsigned int orig_nents;
> +	bool revoked;
> +};
> +
> +static vm_fault_t vfio_pci_dma_buf_fault(struct vm_fault *vmf)
> +{
> +	unsigned long addr = vmf->address - (vmf->pgoff << PAGE_SHIFT);
> +	struct vm_area_struct *vma = vmf->vma;
> +	struct vfio_pci_dma_buf *priv = vma->vm_private_data;
> +	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
> +	unsigned long pfn, i, j;
> +	phys_addr_t phys;
> +	size_t offset;
> +
> +	if (priv->revoked)
> +		return VM_FAULT_SIGBUS;

->revoked seems to be protected by dma_resv_lock, except not here.

> +
> +	down_read(&priv->vdev->memory_lock);
> +
> +	for (i = 0, j = 0; i < priv->nr_ranges && j < vma_pages(vma); i++) {
> +		phys = pci_resource_start(priv->vdev->pdev,
> +					  dma_ranges[i].region_index);
> +		phys += dma_ranges[i].offset;
> +
> +		for (offset = 0; offset != dma_ranges[i].length;) {
> +			pfn = (phys + offset) >> PAGE_SHIFT;
> +
> +			if (vmf_insert_pfn(vma, addr, pfn) != VM_FAULT_NOPAGE) {
> +				up_read(&priv->vdev->memory_lock);
> +				return VM_FAULT_SIGBUS;
> +			}
> +
> +			addr += PAGE_SIZE;
> +			offset += PAGE_SIZE;
> +			if (++j == vma_pages(vma))
> +				break;
> +		}
> +	}
> +
> +	up_read(&priv->vdev->memory_lock);
> +
> +	return VM_FAULT_NOPAGE;
> +}
> +
> +static const struct vm_operations_struct vfio_pci_dma_buf_vmops = {
> +	.fault = vfio_pci_dma_buf_fault,
> +};
> +
> +static int vfio_pci_dma_buf_mmap(struct dma_buf *dmabuf,
> +				 struct vm_area_struct *vma)
> +{
> +	struct vfio_pci_dma_buf *priv = dmabuf->priv;
> +	struct vfio_pci_core_device *vdev = priv->vdev;
> +	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
> +	int i;
> +
> +	if ((vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) == 0)
> +		return -EINVAL;
> +
> +	for (i = 0; i < priv->nr_ranges; i++)
> +		if (!vdev->bar_mmap_supported[dma_ranges[i].region_index])
> +			return -EINVAL;
> +
> +	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
> +	vma->vm_ops = &vfio_pci_dma_buf_vmops;
> +	vma->vm_private_data = priv;
> +	vm_flags_set(vma, VM_ALLOW_ANY_UNCACHED | VM_IO | VM_PFNMAP |
> +		     VM_DONTEXPAND | VM_DONTDUMP);
> +	return 0;
> +}
> +
> +static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
> +				   struct dma_buf_attachment *attachment)
> +{
> +	struct vfio_pci_dma_buf *priv = dmabuf->priv;
> +	int rc;
> +
> +	rc = pci_p2pdma_distance_many(priv->vdev->pdev, &attachment->dev, 1,
> +				      true);
> +	if (rc < 0)
> +		attachment->peer2peer = false;
> +	return 0;
> +}
> +
> +static void vfio_pci_dma_buf_unpin(struct dma_buf_attachment *attachment)
> +{
> +}
> +
> +static int vfio_pci_dma_buf_pin(struct dma_buf_attachment *attachment)
> +{
> +	/*
> +	 * Uses the dynamic interface but must always allow for
> +	 * dma_buf_move_notify() to do revoke
> +	 */
> +	return -EINVAL;
> +}
> +
> +static int populate_sgt(struct dma_buf_attachment *attachment,
> +			enum dma_data_direction dir,
> +			struct sg_table *sgt, size_t sgl_size)
> +{
> +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> +	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
> +	size_t offset, chunk_size;
> +	struct scatterlist *sgl;
> +	dma_addr_t dma_addr;
> +	phys_addr_t phys;
> +	int i, j, ret;
> +
> +	for_each_sgtable_sg(sgt, sgl, j)
> +		sgl->length = 0;
> +
> +	sgl = sgt->sgl;
> +	for (i = 0; i < priv->nr_ranges; i++) {
> +		phys = pci_resource_start(priv->vdev->pdev,
> +					  dma_ranges[i].region_index);
> +		phys += dma_ranges[i].offset;
> +
> +		/*
> +		 * Break the BAR's physical range up into max sized SGL's
> +		 * according to the device's requirement.
> +		 */
> +		for (offset = 0; offset != dma_ranges[i].length;) {
> +			chunk_size = min(dma_ranges[i].length - offset,
> +					 sgl_size);
> +
> +			/*
> +			 * Since the memory being mapped is a device memory
> +			 * it could never be in CPU caches.
> +			 */
> +			dma_addr = dma_map_resource(attachment->dev,
> +						    phys + offset,
> +						    chunk_size, dir,
> +						    DMA_ATTR_SKIP_CPU_SYNC);
> +			ret = dma_mapping_error(attachment->dev, dma_addr);
> +			if (ret)
> +				goto err;
> +
> +			sg_set_page(sgl, NULL, chunk_size, 0);
> +			sg_dma_address(sgl) = dma_addr;
> +			sg_dma_len(sgl) = chunk_size;
> +			sgl = sg_next(sgl);
> +			offset += chunk_size;
> +		}
> +	}
> +
> +	return 0;
> +err:
> +	for_each_sgtable_sg(sgt, sgl, j) {
> +		if (!sg_dma_len(sgl))
> +			continue;
> +
> +		dma_unmap_resource(attachment->dev, sg_dma_address(sgl),
> +				   sg_dma_len(sgl),
> +				   dir, DMA_ATTR_SKIP_CPU_SYNC);
> +	}
> +
> +	return ret;
> +}
> +
> +static struct sg_table *
> +vfio_pci_dma_buf_map(struct dma_buf_attachment *attachment,
> +		     enum dma_data_direction dir)
> +{
> +	size_t sgl_size = dma_get_max_seg_size(attachment->dev);
> +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> +	struct sg_table *sgt;
> +	unsigned int nents;
> +	int ret;
> +
> +	dma_resv_assert_held(priv->dmabuf->resv);
> +
> +	if (!attachment->peer2peer)
> +		return ERR_PTR(-EPERM);
> +
> +	if (priv->revoked)
> +		return ERR_PTR(-ENODEV);

I think a doc patch to dma_buf_ops.map_dma_buf that ENODEV signals
terminal revocation would be good. That patch would also be a really good
place to collected Acks from all the stakeholders that they're fine with
this change, since thus far map_dma_buf pretty much only failed when we've
run out of (device) memory.

> +
> +	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
> +	if (!sgt)
> +		return ERR_PTR(-ENOMEM);
> +
> +	nents = DIV_ROUND_UP(priv->dmabuf->size, sgl_size);
> +	ret = sg_alloc_table(sgt, nents, GFP_KERNEL);
> +	if (ret)
> +		goto err_kfree_sgt;
> +
> +	ret = populate_sgt(attachment, dir, sgt, sgl_size);
> +	if (ret)
> +		goto err_free_sgt;
> +
> +	/*
> +	 * Because we are not going to include a CPU list we want to have some
> +	 * chance that other users will detect this by setting the orig_nents to
> +	 * 0 and using only nents (length of DMA list) when going over the sgl
> +	 */
> +	priv->orig_nents = sgt->orig_nents;
> +	sgt->orig_nents = 0;
> +	return sgt;
> +
> +err_free_sgt:
> +	sg_free_table(sgt);
> +err_kfree_sgt:
> +	kfree(sgt);
> +	return ERR_PTR(ret);
> +}
> +
> +static void vfio_pci_dma_buf_unmap(struct dma_buf_attachment *attachment,
> +				   struct sg_table *sgt,
> +				   enum dma_data_direction dir)
> +{
> +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> +	struct scatterlist *sgl;
> +	int i;
> +
> +	for_each_sgtable_dma_sg(sgt, sgl, i)
> +		dma_unmap_resource(attachment->dev,
> +				   sg_dma_address(sgl),
> +				   sg_dma_len(sgl),
> +				   dir, DMA_ATTR_SKIP_CPU_SYNC);
> +
> +	sgt->orig_nents = priv->orig_nents;
> +	sg_free_table(sgt);
> +	kfree(sgt);
> +}
> +
> +static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf)
> +{
> +	struct vfio_pci_dma_buf *priv = dmabuf->priv;
> +
> +	/*
> +	 * Either this or vfio_pci_dma_buf_cleanup() will remove from the list.
> +	 * The refcount prevents both.
> +	 */
> +	if (priv->vdev) {
> +		down_write(&priv->vdev->memory_lock);
> +		list_del_init(&priv->dmabufs_elm);
> +		up_write(&priv->vdev->memory_lock);
> +		vfio_device_put_registration(&priv->vdev->vdev);
> +	}
> +	kfree(priv);
> +}
> +
> +static const struct dma_buf_ops vfio_pci_dmabuf_ops = {
> +	.attach = vfio_pci_dma_buf_attach,
> +	.map_dma_buf = vfio_pci_dma_buf_map,
> +	.pin = vfio_pci_dma_buf_pin,
> +	.unpin = vfio_pci_dma_buf_unpin,
> +	.release = vfio_pci_dma_buf_release,
> +	.unmap_dma_buf = vfio_pci_dma_buf_unmap,
> +	.mmap = vfio_pci_dma_buf_mmap,
> +};
> +
> +static int check_dma_ranges(struct vfio_pci_dma_buf *priv,
> +			    uint64_t *dmabuf_size)
> +{
> +	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
> +	struct pci_dev *pdev = priv->vdev->pdev;
> +	resource_size_t bar_size;
> +	int i;
> +
> +	for (i = 0; i < priv->nr_ranges; i++) {
> +		/*
> +		 * For PCI the region_index is the BAR number like
> +		 * everything else.
> +		 */
> +		if (dma_ranges[i].region_index >= VFIO_PCI_ROM_REGION_INDEX)
> +			return -EINVAL;
> +
> +		if (!IS_ALIGNED(dma_ranges[i].offset, PAGE_SIZE) ||
> +		    !IS_ALIGNED(dma_ranges[i].length, PAGE_SIZE))
> +			return -EINVAL;
> +
> +		bar_size = pci_resource_len(pdev, dma_ranges[i].region_index);
> +		if (dma_ranges[i].offset > bar_size ||
> +		    dma_ranges[i].offset + dma_ranges[i].length > bar_size)
> +			return -EINVAL;
> +
> +		*dmabuf_size += dma_ranges[i].length;
> +	}
> +
> +	return 0;
> +}
> +
> +int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
> +				  struct vfio_device_feature_dma_buf __user *arg,
> +				  size_t argsz)
> +{
> +	struct vfio_device_feature_dma_buf get_dma_buf;
> +	struct vfio_region_dma_range *dma_ranges;
> +	DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
> +	struct vfio_pci_dma_buf *priv;
> +	uint64_t dmabuf_size = 0;
> +	int ret;
> +
> +	ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_GET,
> +				 sizeof(get_dma_buf));
> +	if (ret != 1)
> +		return ret;
> +
> +	if (copy_from_user(&get_dma_buf, arg, sizeof(get_dma_buf)))
> +		return -EFAULT;
> +
> +	dma_ranges = memdup_array_user(&arg->dma_ranges,
> +				      get_dma_buf.nr_ranges,
> +				      sizeof(*dma_ranges));
> +	if (IS_ERR(dma_ranges))
> +		return PTR_ERR(dma_ranges);
> +
> +	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
> +	if (!priv) {
> +		kfree(dma_ranges);
> +		return -ENOMEM;
> +	}
> +
> +	priv->vdev = vdev;
> +	priv->nr_ranges = get_dma_buf.nr_ranges;
> +	priv->dma_ranges = dma_ranges;
> +
> +	ret = check_dma_ranges(priv, &dmabuf_size);
> +	if (ret)
> +		goto err_free_priv;
> +
> +	if (!vfio_device_try_get_registration(&vdev->vdev)) {
> +		ret = -ENODEV;
> +		goto err_free_priv;
> +	}
> +
> +	exp_info.ops = &vfio_pci_dmabuf_ops;
> +	exp_info.size = dmabuf_size;
> +	exp_info.flags = get_dma_buf.open_flags;
> +	exp_info.priv = priv;
> +
> +	priv->dmabuf = dma_buf_export(&exp_info);
> +	if (IS_ERR(priv->dmabuf)) {
> +		ret = PTR_ERR(priv->dmabuf);
> +		goto err_dev_put;
> +	}
> +
> +	/* dma_buf_put() now frees priv */
> +	INIT_LIST_HEAD(&priv->dmabufs_elm);
> +	down_write(&vdev->memory_lock);
> +	dma_resv_lock(priv->dmabuf->resv, NULL);
> +	priv->revoked = !__vfio_pci_memory_enabled(vdev);
> +	list_add_tail(&priv->dmabufs_elm, &vdev->dmabufs);
> +	dma_resv_unlock(priv->dmabuf->resv);
> +	up_write(&vdev->memory_lock);
> +
> +	/*
> +	 * dma_buf_fd() consumes the reference, when the file closes the dmabuf
> +	 * will be released.
> +	 */
> +	return dma_buf_fd(priv->dmabuf, get_dma_buf.open_flags);
> +
> +err_dev_put:
> +	vfio_device_put_registration(&vdev->vdev);
> +err_free_priv:
> +	kfree(dma_ranges);
> +	kfree(priv);
> +	return ret;
> +}
> +
> +static void revoke_mmap_mappings(struct vfio_pci_dma_buf *priv)
> +{
> +	struct inode *inode = file_inode(priv->dmabuf->file);
> +
> +	unmap_mapping_range(inode->i_mapping, 0, priv->dmabuf->size, true);

I was freaking out for a moment until I remembered we've long ago switched
over to per-dma_buf inodes :-)

> +}
> +
> +void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
> +{
> +	struct vfio_pci_dma_buf *priv;
> +	struct vfio_pci_dma_buf *tmp;
> +
> +	lockdep_assert_held_write(&vdev->memory_lock);
> +
> +	list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
> +		/*
> +		 * Returns true if a reference was successfully obtained.
> +		 * The caller must interlock with the dmabuf's release
> +		 * function in some way, such as RCU, to ensure that this

s/some way/by holding vdev->memory_lock/ because as-is this sounds a bit
too much like yolo locking ...

> +		 * is not called on freed memory.
> +		 */
> +		if (!get_file_rcu(&priv->dmabuf->file))
> +			continue;
> +
> +		if (priv->revoked != revoked) {
> +			dma_resv_lock(priv->dmabuf->resv, NULL);
> +			priv->revoked = revoked;
> +			dma_buf_move_notify(priv->dmabuf);
> +
> +			if (revoked)
> +				revoke_mmap_mappings(priv);
> +
> +			dma_resv_unlock(priv->dmabuf->resv);
> +		}
> +		dma_buf_put(priv->dmabuf);
> +	}
> +}
> +
> +void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
> +{
> +	struct vfio_pci_dma_buf *priv;
> +	struct vfio_pci_dma_buf *tmp;
> +
> +	down_write(&vdev->memory_lock);
> +	list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
> +		if (!get_file_rcu(&priv->dmabuf->file))
> +			continue;
> +		dma_resv_lock(priv->dmabuf->resv, NULL);
> +		list_del_init(&priv->dmabufs_elm);
> +		priv->vdev = NULL;
> +		priv->revoked = true;
> +		dma_buf_move_notify(priv->dmabuf);
> +		revoke_mmap_mappings(priv);
> +		dma_resv_unlock(priv->dmabuf->resv);
> +		vfio_device_put_registration(&vdev->vdev);
> +		dma_buf_put(priv->dmabuf);
> +	}
> +	up_write(&vdev->memory_lock);
> +}
> diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
> index 97422aafaa7b..5cd7ec4b5e40 100644
> --- a/drivers/vfio/pci/vfio_pci_config.c
> +++ b/drivers/vfio/pci/vfio_pci_config.c
> @@ -585,10 +585,12 @@ static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos,
>  		virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
>  		new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
>  
> -		if (!new_mem)
> +		if (!new_mem) {
>  			vfio_pci_zap_and_down_write_memory_lock(vdev);
> -		else
> +			vfio_pci_dma_buf_move(vdev, true);
> +		} else {
>  			down_write(&vdev->memory_lock);
> +		}
>  
>  		/*
>  		 * If the user is writing mem/io enable (new_mem/io) and we
> @@ -623,6 +625,8 @@ static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos,
>  		*virt_cmd &= cpu_to_le16(~mask);
>  		*virt_cmd |= cpu_to_le16(new_cmd & mask);
>  
> +		if (__vfio_pci_memory_enabled(vdev))
> +			vfio_pci_dma_buf_move(vdev, false);
>  		up_write(&vdev->memory_lock);
>  	}
>  
> @@ -703,12 +707,16 @@ static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
>  static void vfio_lock_and_set_power_state(struct vfio_pci_core_device *vdev,
>  					  pci_power_t state)
>  {
> -	if (state >= PCI_D3hot)
> +	if (state >= PCI_D3hot) {
>  		vfio_pci_zap_and_down_write_memory_lock(vdev);
> -	else
> +		vfio_pci_dma_buf_move(vdev, true);
> +	} else {
>  		down_write(&vdev->memory_lock);
> +	}
>  
>  	vfio_pci_set_power_state(vdev, state);
> +	if (__vfio_pci_memory_enabled(vdev))
> +		vfio_pci_dma_buf_move(vdev, false);
>  	up_write(&vdev->memory_lock);
>  }
>  
> @@ -896,7 +904,10 @@ static int vfio_exp_config_write(struct vfio_pci_core_device *vdev, int pos,
>  
>  		if (!ret && (cap & PCI_EXP_DEVCAP_FLR)) {
>  			vfio_pci_zap_and_down_write_memory_lock(vdev);
> +			vfio_pci_dma_buf_move(vdev, true);
>  			pci_try_reset_function(vdev->pdev);
> +			if (__vfio_pci_memory_enabled(vdev))
> +				vfio_pci_dma_buf_move(vdev, true);
>  			up_write(&vdev->memory_lock);
>  		}
>  	}
> @@ -978,7 +989,10 @@ static int vfio_af_config_write(struct vfio_pci_core_device *vdev, int pos,
>  
>  		if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP)) {
>  			vfio_pci_zap_and_down_write_memory_lock(vdev);
> +			vfio_pci_dma_buf_move(vdev, true);
>  			pci_try_reset_function(vdev->pdev);
> +			if (__vfio_pci_memory_enabled(vdev))
> +				vfio_pci_dma_buf_move(vdev, true);
>  			up_write(&vdev->memory_lock);
>  		}
>  	}
> diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
> index 3282ef2dddea..154129133d5e 100644
> --- a/drivers/vfio/pci/vfio_pci_core.c
> +++ b/drivers/vfio/pci/vfio_pci_core.c
> @@ -291,6 +291,8 @@ static int vfio_pci_runtime_pm_entry(struct vfio_pci_core_device *vdev,
>  	 * semaphore.
>  	 */
>  	vfio_pci_zap_and_down_write_memory_lock(vdev);
> +	vfio_pci_dma_buf_move(vdev, true);
> +
>  	if (vdev->pm_runtime_engaged) {
>  		up_write(&vdev->memory_lock);
>  		return -EINVAL;
> @@ -374,6 +376,8 @@ static void vfio_pci_runtime_pm_exit(struct vfio_pci_core_device *vdev)
>  	 */
>  	down_write(&vdev->memory_lock);
>  	__vfio_pci_runtime_pm_exit(vdev);
> +	if (__vfio_pci_memory_enabled(vdev))
> +		vfio_pci_dma_buf_move(vdev, false);
>  	up_write(&vdev->memory_lock);
>  }
>  
> @@ -694,6 +698,8 @@ void vfio_pci_core_close_device(struct vfio_device *core_vdev)
>  #endif
>  	vfio_pci_core_disable(vdev);
>  
> +	vfio_pci_dma_buf_cleanup(vdev);
> +
>  	mutex_lock(&vdev->igate);
>  	if (vdev->err_trigger) {
>  		eventfd_ctx_put(vdev->err_trigger);
> @@ -1238,7 +1244,10 @@ static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev,
>  	 */
>  	vfio_pci_set_power_state(vdev, PCI_D0);
>  
> +	vfio_pci_dma_buf_move(vdev, true);
>  	ret = pci_try_reset_function(vdev->pdev);
> +	if (__vfio_pci_memory_enabled(vdev))
> +		vfio_pci_dma_buf_move(vdev, false);
>  	up_write(&vdev->memory_lock);
>  
>  	return ret;
> @@ -1527,6 +1536,8 @@ int vfio_pci_core_ioctl_feature(struct vfio_device *device, u32 flags,
>  		return vfio_pci_core_pm_exit(vdev, flags, arg, argsz);
>  	case VFIO_DEVICE_FEATURE_PCI_VF_TOKEN:
>  		return vfio_pci_core_feature_token(vdev, flags, arg, argsz);
> +	case VFIO_DEVICE_FEATURE_DMA_BUF:
> +		return vfio_pci_core_feature_dma_buf(vdev, flags, arg, argsz);
>  	default:
>  		return -ENOTTY;
>  	}
> @@ -2077,6 +2088,7 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev)
>  	INIT_LIST_HEAD(&vdev->dummy_resources_list);
>  	INIT_LIST_HEAD(&vdev->ioeventfds_list);
>  	INIT_LIST_HEAD(&vdev->sriov_pfs_item);
> +	INIT_LIST_HEAD(&vdev->dmabufs);
>  	init_rwsem(&vdev->memory_lock);
>  	xa_init(&vdev->ctx);
>  
> @@ -2459,11 +2471,17 @@ static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set,
>  	 * cause the PCI config space reset without restoring the original
>  	 * state (saved locally in 'vdev->pm_save').
>  	 */
> -	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list)
> +	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list) {
> +		vfio_pci_dma_buf_move(vdev, true);
>  		vfio_pci_set_power_state(vdev, PCI_D0);
> +	}
>  
>  	ret = pci_reset_bus(pdev);
>  
> +	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list)
> +		if (__vfio_pci_memory_enabled(vdev))
> +			vfio_pci_dma_buf_move(vdev, false);
> +
>  	vdev = list_last_entry(&dev_set->device_list,
>  			       struct vfio_pci_core_device, vdev.dev_set_list);
>  
> diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h
> index 5e4fa69aee16..09d3c300918c 100644
> --- a/drivers/vfio/pci/vfio_pci_priv.h
> +++ b/drivers/vfio/pci/vfio_pci_priv.h
> @@ -101,4 +101,27 @@ static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
>  	return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
>  }
>  
> +#ifdef CONFIG_DMA_SHARED_BUFFER
> +int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
> +				  struct vfio_device_feature_dma_buf __user *arg,
> +				  size_t argsz);
> +void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev);
> +void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked);
> +#else
> +static int
> +vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
> +			      struct vfio_device_feature_dma_buf __user *arg,
> +			      size_t argsz)
> +{
> +	return -ENOTTY;
> +}
> +static inline void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
> +{
> +}
> +static inline void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev,
> +					 bool revoked)
> +{
> +}
> +#endif
> +
>  #endif
> diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h
> index f87067438ed4..3756e0904de2 100644
> --- a/include/linux/vfio_pci_core.h
> +++ b/include/linux/vfio_pci_core.h
> @@ -94,6 +94,7 @@ struct vfio_pci_core_device {
>  	struct vfio_pci_core_device	*sriov_pf_core_dev;
>  	struct notifier_block	nb;
>  	struct rw_semaphore	memory_lock;
> +	struct list_head	dmabufs;
>  };
>  
>  /* Will be exported for vfio pci drivers usage */
> diff --git a/include/uapi/linux/vfio.h b/include/uapi/linux/vfio.h
> index 2b68e6cdf190..6f283dadacb7 100644
> --- a/include/uapi/linux/vfio.h
> +++ b/include/uapi/linux/vfio.h
> @@ -1458,6 +1458,31 @@ struct vfio_device_feature_bus_master {
>  };
>  #define VFIO_DEVICE_FEATURE_BUS_MASTER 10
>  
> +/**
> + * Upon VFIO_DEVICE_FEATURE_GET create a dma_buf fd for the
> + * regions selected.
> + *
> + * open_flags are the typical flags passed to open(2), eg O_RDWR, O_CLOEXEC,
> + * etc. offset/length specify a slice of the region to create the dmabuf from.
> + * nr_ranges is the total number of (P2P DMA) ranges that comprise the dmabuf.
> + *
> + * Return: The fd number on success, -1 and errno is set on failure.
> + */
> +#define VFIO_DEVICE_FEATURE_DMA_BUF 11
> +
> +struct vfio_region_dma_range {
> +	__u32	region_index;
> +	__u32	__pad;
> +	__u64	offset;
> +	__u64	length;
> +};
> +
> +struct vfio_device_feature_dma_buf {
> +	__u32	open_flags;
> +	__u32	nr_ranges;
> +	struct vfio_region_dma_range dma_ranges[];
> +};
> +
>  /* -------- API for Type1 VFIO IOMMU -------- */
>  
>  /**
> -- 
> 2.45.1
>
Leon Romanovsky June 24, 2024, 2:11 p.m. UTC | #2
On Sun, Jun 23, 2024 at 11:53:11PM -0700, Vivek Kasireddy wrote:
> From Jason Gunthorpe:
> "dma-buf has become a way to safely acquire a handle to non-struct page
> memory that can still have lifetime controlled by the exporter. Notably
> RDMA can now import dma-buf FDs and build them into MRs which allows for
> PCI P2P operations. Extend this to allow vfio-pci to export MMIO memory
> from PCI device BARs.
> 
> The patch design loosely follows the pattern in commit
> db1a8dd916aa ("habanalabs: add support for dma-buf exporter") except this
> does not support pinning.
> 
> Instead, this implements what, in the past, we've called a revocable
> attachment using move. In normal situations the attachment is pinned, as a
> BAR does not change physical address. However when the VFIO device is
> closed, or a PCI reset is issued, access to the MMIO memory is revoked.
> 
> Revoked means that move occurs, but an attempt to immediately re-map the
> memory will fail. In the reset case a future move will be triggered when
> MMIO access returns. As both close and reset are under userspace control
> it is expected that userspace will suspend use of the dma-buf before doing
> these operations, the revoke is purely for kernel self-defense against a
> hostile userspace."
> 
> Following enhancements are made to the original patch:
> - Add support for creating dmabuf from multiple areas (or ranges)
> - Add a mmap handler to provide CPU access to the dmabuf
> 
> Original-patch-by: Jason Gunthorpe <jgg@nvidia.com>
> Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
> ---
>  drivers/vfio/pci/Makefile          |   1 +
>  drivers/vfio/pci/dma_buf.c         | 438 +++++++++++++++++++++++++++++
>  drivers/vfio/pci/vfio_pci_config.c |  22 +-
>  drivers/vfio/pci/vfio_pci_core.c   |  20 +-
>  drivers/vfio/pci/vfio_pci_priv.h   |  23 ++
>  include/linux/vfio_pci_core.h      |   1 +
>  include/uapi/linux/vfio.h          |  25 ++
>  7 files changed, 525 insertions(+), 5 deletions(-)
>  create mode 100644 drivers/vfio/pci/dma_buf.c

<...>

> +static int populate_sgt(struct dma_buf_attachment *attachment,
> +			enum dma_data_direction dir,
> +			struct sg_table *sgt, size_t sgl_size)
> +{
> +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> +	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
> +	size_t offset, chunk_size;
> +	struct scatterlist *sgl;
> +	dma_addr_t dma_addr;
> +	phys_addr_t phys;
> +	int i, j, ret;
> +
> +	for_each_sgtable_sg(sgt, sgl, j)
> +		sgl->length = 0;
> +
> +	sgl = sgt->sgl;
> +	for (i = 0; i < priv->nr_ranges; i++) {
> +		phys = pci_resource_start(priv->vdev->pdev,
> +					  dma_ranges[i].region_index);
> +		phys += dma_ranges[i].offset;
> +
> +		/*
> +		 * Break the BAR's physical range up into max sized SGL's
> +		 * according to the device's requirement.
> +		 */
> +		for (offset = 0; offset != dma_ranges[i].length;) {
> +			chunk_size = min(dma_ranges[i].length - offset,
> +					 sgl_size);
> +
> +			/*
> +			 * Since the memory being mapped is a device memory
> +			 * it could never be in CPU caches.
> +			 */
> +			dma_addr = dma_map_resource(attachment->dev,
> +						    phys + offset,
> +						    chunk_size, dir,
> +						    DMA_ATTR_SKIP_CPU_SYNC);
> +			ret = dma_mapping_error(attachment->dev, dma_addr);
> +			if (ret)
> +				goto err;
> +
> +			sg_set_page(sgl, NULL, chunk_size, 0);
> +			sg_dma_address(sgl) = dma_addr;
> +			sg_dma_len(sgl) = chunk_size;
> +			sgl = sg_next(sgl);
> +			offset += chunk_size;
> +		}
> +	}
> +
> +	return 0;
> +err:
> +	for_each_sgtable_sg(sgt, sgl, j) {
> +		if (!sg_dma_len(sgl))
> +			continue;
> +
> +		dma_unmap_resource(attachment->dev, sg_dma_address(sgl),
> +				   sg_dma_len(sgl),
> +				   dir, DMA_ATTR_SKIP_CPU_SYNC);
> +	}
> +
> +	return ret;
> +}
> +
> +static struct sg_table *
> +vfio_pci_dma_buf_map(struct dma_buf_attachment *attachment,
> +		     enum dma_data_direction dir)
> +{
> +	size_t sgl_size = dma_get_max_seg_size(attachment->dev);
> +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> +	struct sg_table *sgt;
> +	unsigned int nents;
> +	int ret;
> +
> +	dma_resv_assert_held(priv->dmabuf->resv);
> +
> +	if (!attachment->peer2peer)
> +		return ERR_PTR(-EPERM);
> +
> +	if (priv->revoked)
> +		return ERR_PTR(-ENODEV);
> +
> +	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
> +	if (!sgt)
> +		return ERR_PTR(-ENOMEM);
> +
> +	nents = DIV_ROUND_UP(priv->dmabuf->size, sgl_size);
> +	ret = sg_alloc_table(sgt, nents, GFP_KERNEL);
> +	if (ret)
> +		goto err_kfree_sgt;
> +
> +	ret = populate_sgt(attachment, dir, sgt, sgl_size);

One of the outcomes of the discussion over original Jason's series was
the decision do not use scatter-gather list, but provide DMA API
to be usable for non-struct page memory, which eliminates the need
of intermediate SG list.

This is why we had this series https://lore.kernel.org/all/cover.1709635535.git.leon@kernel.org

And new version is available here:
https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=dma-split-v1

Thanks
diff mbox series

Patch

diff --git a/drivers/vfio/pci/Makefile b/drivers/vfio/pci/Makefile
index cf00c0a7e55c..0cfdc9ede82f 100644
--- a/drivers/vfio/pci/Makefile
+++ b/drivers/vfio/pci/Makefile
@@ -2,6 +2,7 @@ 
 
 vfio-pci-core-y := vfio_pci_core.o vfio_pci_intrs.o vfio_pci_rdwr.o vfio_pci_config.o
 vfio-pci-core-$(CONFIG_VFIO_PCI_ZDEV_KVM) += vfio_pci_zdev.o
+vfio-pci-core-$(CONFIG_DMA_SHARED_BUFFER) += dma_buf.o
 obj-$(CONFIG_VFIO_PCI_CORE) += vfio-pci-core.o
 
 vfio-pci-y := vfio_pci.o
diff --git a/drivers/vfio/pci/dma_buf.c b/drivers/vfio/pci/dma_buf.c
new file mode 100644
index 000000000000..68760677c6a0
--- /dev/null
+++ b/drivers/vfio/pci/dma_buf.c
@@ -0,0 +1,438 @@ 
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES.
+ */
+#include <linux/dma-buf.h>
+#include <linux/pci-p2pdma.h>
+#include <linux/dma-resv.h>
+
+#include "vfio_pci_priv.h"
+
+MODULE_IMPORT_NS(DMA_BUF);
+
+struct vfio_pci_dma_buf {
+	struct dma_buf *dmabuf;
+	struct vfio_pci_core_device *vdev;
+	struct list_head dmabufs_elm;
+	unsigned int nr_ranges;
+	struct vfio_region_dma_range *dma_ranges;
+	unsigned int orig_nents;
+	bool revoked;
+};
+
+static vm_fault_t vfio_pci_dma_buf_fault(struct vm_fault *vmf)
+{
+	unsigned long addr = vmf->address - (vmf->pgoff << PAGE_SHIFT);
+	struct vm_area_struct *vma = vmf->vma;
+	struct vfio_pci_dma_buf *priv = vma->vm_private_data;
+	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
+	unsigned long pfn, i, j;
+	phys_addr_t phys;
+	size_t offset;
+
+	if (priv->revoked)
+		return VM_FAULT_SIGBUS;
+
+	down_read(&priv->vdev->memory_lock);
+
+	for (i = 0, j = 0; i < priv->nr_ranges && j < vma_pages(vma); i++) {
+		phys = pci_resource_start(priv->vdev->pdev,
+					  dma_ranges[i].region_index);
+		phys += dma_ranges[i].offset;
+
+		for (offset = 0; offset != dma_ranges[i].length;) {
+			pfn = (phys + offset) >> PAGE_SHIFT;
+
+			if (vmf_insert_pfn(vma, addr, pfn) != VM_FAULT_NOPAGE) {
+				up_read(&priv->vdev->memory_lock);
+				return VM_FAULT_SIGBUS;
+			}
+
+			addr += PAGE_SIZE;
+			offset += PAGE_SIZE;
+			if (++j == vma_pages(vma))
+				break;
+		}
+	}
+
+	up_read(&priv->vdev->memory_lock);
+
+	return VM_FAULT_NOPAGE;
+}
+
+static const struct vm_operations_struct vfio_pci_dma_buf_vmops = {
+	.fault = vfio_pci_dma_buf_fault,
+};
+
+static int vfio_pci_dma_buf_mmap(struct dma_buf *dmabuf,
+				 struct vm_area_struct *vma)
+{
+	struct vfio_pci_dma_buf *priv = dmabuf->priv;
+	struct vfio_pci_core_device *vdev = priv->vdev;
+	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
+	int i;
+
+	if ((vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) == 0)
+		return -EINVAL;
+
+	for (i = 0; i < priv->nr_ranges; i++)
+		if (!vdev->bar_mmap_supported[dma_ranges[i].region_index])
+			return -EINVAL;
+
+	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+	vma->vm_ops = &vfio_pci_dma_buf_vmops;
+	vma->vm_private_data = priv;
+	vm_flags_set(vma, VM_ALLOW_ANY_UNCACHED | VM_IO | VM_PFNMAP |
+		     VM_DONTEXPAND | VM_DONTDUMP);
+	return 0;
+}
+
+static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
+				   struct dma_buf_attachment *attachment)
+{
+	struct vfio_pci_dma_buf *priv = dmabuf->priv;
+	int rc;
+
+	rc = pci_p2pdma_distance_many(priv->vdev->pdev, &attachment->dev, 1,
+				      true);
+	if (rc < 0)
+		attachment->peer2peer = false;
+	return 0;
+}
+
+static void vfio_pci_dma_buf_unpin(struct dma_buf_attachment *attachment)
+{
+}
+
+static int vfio_pci_dma_buf_pin(struct dma_buf_attachment *attachment)
+{
+	/*
+	 * Uses the dynamic interface but must always allow for
+	 * dma_buf_move_notify() to do revoke
+	 */
+	return -EINVAL;
+}
+
+static int populate_sgt(struct dma_buf_attachment *attachment,
+			enum dma_data_direction dir,
+			struct sg_table *sgt, size_t sgl_size)
+{
+	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
+	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
+	size_t offset, chunk_size;
+	struct scatterlist *sgl;
+	dma_addr_t dma_addr;
+	phys_addr_t phys;
+	int i, j, ret;
+
+	for_each_sgtable_sg(sgt, sgl, j)
+		sgl->length = 0;
+
+	sgl = sgt->sgl;
+	for (i = 0; i < priv->nr_ranges; i++) {
+		phys = pci_resource_start(priv->vdev->pdev,
+					  dma_ranges[i].region_index);
+		phys += dma_ranges[i].offset;
+
+		/*
+		 * Break the BAR's physical range up into max sized SGL's
+		 * according to the device's requirement.
+		 */
+		for (offset = 0; offset != dma_ranges[i].length;) {
+			chunk_size = min(dma_ranges[i].length - offset,
+					 sgl_size);
+
+			/*
+			 * Since the memory being mapped is a device memory
+			 * it could never be in CPU caches.
+			 */
+			dma_addr = dma_map_resource(attachment->dev,
+						    phys + offset,
+						    chunk_size, dir,
+						    DMA_ATTR_SKIP_CPU_SYNC);
+			ret = dma_mapping_error(attachment->dev, dma_addr);
+			if (ret)
+				goto err;
+
+			sg_set_page(sgl, NULL, chunk_size, 0);
+			sg_dma_address(sgl) = dma_addr;
+			sg_dma_len(sgl) = chunk_size;
+			sgl = sg_next(sgl);
+			offset += chunk_size;
+		}
+	}
+
+	return 0;
+err:
+	for_each_sgtable_sg(sgt, sgl, j) {
+		if (!sg_dma_len(sgl))
+			continue;
+
+		dma_unmap_resource(attachment->dev, sg_dma_address(sgl),
+				   sg_dma_len(sgl),
+				   dir, DMA_ATTR_SKIP_CPU_SYNC);
+	}
+
+	return ret;
+}
+
+static struct sg_table *
+vfio_pci_dma_buf_map(struct dma_buf_attachment *attachment,
+		     enum dma_data_direction dir)
+{
+	size_t sgl_size = dma_get_max_seg_size(attachment->dev);
+	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
+	struct sg_table *sgt;
+	unsigned int nents;
+	int ret;
+
+	dma_resv_assert_held(priv->dmabuf->resv);
+
+	if (!attachment->peer2peer)
+		return ERR_PTR(-EPERM);
+
+	if (priv->revoked)
+		return ERR_PTR(-ENODEV);
+
+	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
+	if (!sgt)
+		return ERR_PTR(-ENOMEM);
+
+	nents = DIV_ROUND_UP(priv->dmabuf->size, sgl_size);
+	ret = sg_alloc_table(sgt, nents, GFP_KERNEL);
+	if (ret)
+		goto err_kfree_sgt;
+
+	ret = populate_sgt(attachment, dir, sgt, sgl_size);
+	if (ret)
+		goto err_free_sgt;
+
+	/*
+	 * Because we are not going to include a CPU list we want to have some
+	 * chance that other users will detect this by setting the orig_nents to
+	 * 0 and using only nents (length of DMA list) when going over the sgl
+	 */
+	priv->orig_nents = sgt->orig_nents;
+	sgt->orig_nents = 0;
+	return sgt;
+
+err_free_sgt:
+	sg_free_table(sgt);
+err_kfree_sgt:
+	kfree(sgt);
+	return ERR_PTR(ret);
+}
+
+static void vfio_pci_dma_buf_unmap(struct dma_buf_attachment *attachment,
+				   struct sg_table *sgt,
+				   enum dma_data_direction dir)
+{
+	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
+	struct scatterlist *sgl;
+	int i;
+
+	for_each_sgtable_dma_sg(sgt, sgl, i)
+		dma_unmap_resource(attachment->dev,
+				   sg_dma_address(sgl),
+				   sg_dma_len(sgl),
+				   dir, DMA_ATTR_SKIP_CPU_SYNC);
+
+	sgt->orig_nents = priv->orig_nents;
+	sg_free_table(sgt);
+	kfree(sgt);
+}
+
+static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf)
+{
+	struct vfio_pci_dma_buf *priv = dmabuf->priv;
+
+	/*
+	 * Either this or vfio_pci_dma_buf_cleanup() will remove from the list.
+	 * The refcount prevents both.
+	 */
+	if (priv->vdev) {
+		down_write(&priv->vdev->memory_lock);
+		list_del_init(&priv->dmabufs_elm);
+		up_write(&priv->vdev->memory_lock);
+		vfio_device_put_registration(&priv->vdev->vdev);
+	}
+	kfree(priv);
+}
+
+static const struct dma_buf_ops vfio_pci_dmabuf_ops = {
+	.attach = vfio_pci_dma_buf_attach,
+	.map_dma_buf = vfio_pci_dma_buf_map,
+	.pin = vfio_pci_dma_buf_pin,
+	.unpin = vfio_pci_dma_buf_unpin,
+	.release = vfio_pci_dma_buf_release,
+	.unmap_dma_buf = vfio_pci_dma_buf_unmap,
+	.mmap = vfio_pci_dma_buf_mmap,
+};
+
+static int check_dma_ranges(struct vfio_pci_dma_buf *priv,
+			    uint64_t *dmabuf_size)
+{
+	struct vfio_region_dma_range *dma_ranges = priv->dma_ranges;
+	struct pci_dev *pdev = priv->vdev->pdev;
+	resource_size_t bar_size;
+	int i;
+
+	for (i = 0; i < priv->nr_ranges; i++) {
+		/*
+		 * For PCI the region_index is the BAR number like
+		 * everything else.
+		 */
+		if (dma_ranges[i].region_index >= VFIO_PCI_ROM_REGION_INDEX)
+			return -EINVAL;
+
+		if (!IS_ALIGNED(dma_ranges[i].offset, PAGE_SIZE) ||
+		    !IS_ALIGNED(dma_ranges[i].length, PAGE_SIZE))
+			return -EINVAL;
+
+		bar_size = pci_resource_len(pdev, dma_ranges[i].region_index);
+		if (dma_ranges[i].offset > bar_size ||
+		    dma_ranges[i].offset + dma_ranges[i].length > bar_size)
+			return -EINVAL;
+
+		*dmabuf_size += dma_ranges[i].length;
+	}
+
+	return 0;
+}
+
+int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
+				  struct vfio_device_feature_dma_buf __user *arg,
+				  size_t argsz)
+{
+	struct vfio_device_feature_dma_buf get_dma_buf;
+	struct vfio_region_dma_range *dma_ranges;
+	DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
+	struct vfio_pci_dma_buf *priv;
+	uint64_t dmabuf_size = 0;
+	int ret;
+
+	ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_GET,
+				 sizeof(get_dma_buf));
+	if (ret != 1)
+		return ret;
+
+	if (copy_from_user(&get_dma_buf, arg, sizeof(get_dma_buf)))
+		return -EFAULT;
+
+	dma_ranges = memdup_array_user(&arg->dma_ranges,
+				      get_dma_buf.nr_ranges,
+				      sizeof(*dma_ranges));
+	if (IS_ERR(dma_ranges))
+		return PTR_ERR(dma_ranges);
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv) {
+		kfree(dma_ranges);
+		return -ENOMEM;
+	}
+
+	priv->vdev = vdev;
+	priv->nr_ranges = get_dma_buf.nr_ranges;
+	priv->dma_ranges = dma_ranges;
+
+	ret = check_dma_ranges(priv, &dmabuf_size);
+	if (ret)
+		goto err_free_priv;
+
+	if (!vfio_device_try_get_registration(&vdev->vdev)) {
+		ret = -ENODEV;
+		goto err_free_priv;
+	}
+
+	exp_info.ops = &vfio_pci_dmabuf_ops;
+	exp_info.size = dmabuf_size;
+	exp_info.flags = get_dma_buf.open_flags;
+	exp_info.priv = priv;
+
+	priv->dmabuf = dma_buf_export(&exp_info);
+	if (IS_ERR(priv->dmabuf)) {
+		ret = PTR_ERR(priv->dmabuf);
+		goto err_dev_put;
+	}
+
+	/* dma_buf_put() now frees priv */
+	INIT_LIST_HEAD(&priv->dmabufs_elm);
+	down_write(&vdev->memory_lock);
+	dma_resv_lock(priv->dmabuf->resv, NULL);
+	priv->revoked = !__vfio_pci_memory_enabled(vdev);
+	list_add_tail(&priv->dmabufs_elm, &vdev->dmabufs);
+	dma_resv_unlock(priv->dmabuf->resv);
+	up_write(&vdev->memory_lock);
+
+	/*
+	 * dma_buf_fd() consumes the reference, when the file closes the dmabuf
+	 * will be released.
+	 */
+	return dma_buf_fd(priv->dmabuf, get_dma_buf.open_flags);
+
+err_dev_put:
+	vfio_device_put_registration(&vdev->vdev);
+err_free_priv:
+	kfree(dma_ranges);
+	kfree(priv);
+	return ret;
+}
+
+static void revoke_mmap_mappings(struct vfio_pci_dma_buf *priv)
+{
+	struct inode *inode = file_inode(priv->dmabuf->file);
+
+	unmap_mapping_range(inode->i_mapping, 0, priv->dmabuf->size, true);
+}
+
+void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
+{
+	struct vfio_pci_dma_buf *priv;
+	struct vfio_pci_dma_buf *tmp;
+
+	lockdep_assert_held_write(&vdev->memory_lock);
+
+	list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
+		/*
+		 * Returns true if a reference was successfully obtained.
+		 * The caller must interlock with the dmabuf's release
+		 * function in some way, such as RCU, to ensure that this
+		 * is not called on freed memory.
+		 */
+		if (!get_file_rcu(&priv->dmabuf->file))
+			continue;
+
+		if (priv->revoked != revoked) {
+			dma_resv_lock(priv->dmabuf->resv, NULL);
+			priv->revoked = revoked;
+			dma_buf_move_notify(priv->dmabuf);
+
+			if (revoked)
+				revoke_mmap_mappings(priv);
+
+			dma_resv_unlock(priv->dmabuf->resv);
+		}
+		dma_buf_put(priv->dmabuf);
+	}
+}
+
+void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
+{
+	struct vfio_pci_dma_buf *priv;
+	struct vfio_pci_dma_buf *tmp;
+
+	down_write(&vdev->memory_lock);
+	list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
+		if (!get_file_rcu(&priv->dmabuf->file))
+			continue;
+		dma_resv_lock(priv->dmabuf->resv, NULL);
+		list_del_init(&priv->dmabufs_elm);
+		priv->vdev = NULL;
+		priv->revoked = true;
+		dma_buf_move_notify(priv->dmabuf);
+		revoke_mmap_mappings(priv);
+		dma_resv_unlock(priv->dmabuf->resv);
+		vfio_device_put_registration(&vdev->vdev);
+		dma_buf_put(priv->dmabuf);
+	}
+	up_write(&vdev->memory_lock);
+}
diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
index 97422aafaa7b..5cd7ec4b5e40 100644
--- a/drivers/vfio/pci/vfio_pci_config.c
+++ b/drivers/vfio/pci/vfio_pci_config.c
@@ -585,10 +585,12 @@  static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos,
 		virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
 		new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
 
-		if (!new_mem)
+		if (!new_mem) {
 			vfio_pci_zap_and_down_write_memory_lock(vdev);
-		else
+			vfio_pci_dma_buf_move(vdev, true);
+		} else {
 			down_write(&vdev->memory_lock);
+		}
 
 		/*
 		 * If the user is writing mem/io enable (new_mem/io) and we
@@ -623,6 +625,8 @@  static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos,
 		*virt_cmd &= cpu_to_le16(~mask);
 		*virt_cmd |= cpu_to_le16(new_cmd & mask);
 
+		if (__vfio_pci_memory_enabled(vdev))
+			vfio_pci_dma_buf_move(vdev, false);
 		up_write(&vdev->memory_lock);
 	}
 
@@ -703,12 +707,16 @@  static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
 static void vfio_lock_and_set_power_state(struct vfio_pci_core_device *vdev,
 					  pci_power_t state)
 {
-	if (state >= PCI_D3hot)
+	if (state >= PCI_D3hot) {
 		vfio_pci_zap_and_down_write_memory_lock(vdev);
-	else
+		vfio_pci_dma_buf_move(vdev, true);
+	} else {
 		down_write(&vdev->memory_lock);
+	}
 
 	vfio_pci_set_power_state(vdev, state);
+	if (__vfio_pci_memory_enabled(vdev))
+		vfio_pci_dma_buf_move(vdev, false);
 	up_write(&vdev->memory_lock);
 }
 
@@ -896,7 +904,10 @@  static int vfio_exp_config_write(struct vfio_pci_core_device *vdev, int pos,
 
 		if (!ret && (cap & PCI_EXP_DEVCAP_FLR)) {
 			vfio_pci_zap_and_down_write_memory_lock(vdev);
+			vfio_pci_dma_buf_move(vdev, true);
 			pci_try_reset_function(vdev->pdev);
+			if (__vfio_pci_memory_enabled(vdev))
+				vfio_pci_dma_buf_move(vdev, true);
 			up_write(&vdev->memory_lock);
 		}
 	}
@@ -978,7 +989,10 @@  static int vfio_af_config_write(struct vfio_pci_core_device *vdev, int pos,
 
 		if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP)) {
 			vfio_pci_zap_and_down_write_memory_lock(vdev);
+			vfio_pci_dma_buf_move(vdev, true);
 			pci_try_reset_function(vdev->pdev);
+			if (__vfio_pci_memory_enabled(vdev))
+				vfio_pci_dma_buf_move(vdev, true);
 			up_write(&vdev->memory_lock);
 		}
 	}
diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index 3282ef2dddea..154129133d5e 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c
@@ -291,6 +291,8 @@  static int vfio_pci_runtime_pm_entry(struct vfio_pci_core_device *vdev,
 	 * semaphore.
 	 */
 	vfio_pci_zap_and_down_write_memory_lock(vdev);
+	vfio_pci_dma_buf_move(vdev, true);
+
 	if (vdev->pm_runtime_engaged) {
 		up_write(&vdev->memory_lock);
 		return -EINVAL;
@@ -374,6 +376,8 @@  static void vfio_pci_runtime_pm_exit(struct vfio_pci_core_device *vdev)
 	 */
 	down_write(&vdev->memory_lock);
 	__vfio_pci_runtime_pm_exit(vdev);
+	if (__vfio_pci_memory_enabled(vdev))
+		vfio_pci_dma_buf_move(vdev, false);
 	up_write(&vdev->memory_lock);
 }
 
@@ -694,6 +698,8 @@  void vfio_pci_core_close_device(struct vfio_device *core_vdev)
 #endif
 	vfio_pci_core_disable(vdev);
 
+	vfio_pci_dma_buf_cleanup(vdev);
+
 	mutex_lock(&vdev->igate);
 	if (vdev->err_trigger) {
 		eventfd_ctx_put(vdev->err_trigger);
@@ -1238,7 +1244,10 @@  static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev,
 	 */
 	vfio_pci_set_power_state(vdev, PCI_D0);
 
+	vfio_pci_dma_buf_move(vdev, true);
 	ret = pci_try_reset_function(vdev->pdev);
+	if (__vfio_pci_memory_enabled(vdev))
+		vfio_pci_dma_buf_move(vdev, false);
 	up_write(&vdev->memory_lock);
 
 	return ret;
@@ -1527,6 +1536,8 @@  int vfio_pci_core_ioctl_feature(struct vfio_device *device, u32 flags,
 		return vfio_pci_core_pm_exit(vdev, flags, arg, argsz);
 	case VFIO_DEVICE_FEATURE_PCI_VF_TOKEN:
 		return vfio_pci_core_feature_token(vdev, flags, arg, argsz);
+	case VFIO_DEVICE_FEATURE_DMA_BUF:
+		return vfio_pci_core_feature_dma_buf(vdev, flags, arg, argsz);
 	default:
 		return -ENOTTY;
 	}
@@ -2077,6 +2088,7 @@  int vfio_pci_core_init_dev(struct vfio_device *core_vdev)
 	INIT_LIST_HEAD(&vdev->dummy_resources_list);
 	INIT_LIST_HEAD(&vdev->ioeventfds_list);
 	INIT_LIST_HEAD(&vdev->sriov_pfs_item);
+	INIT_LIST_HEAD(&vdev->dmabufs);
 	init_rwsem(&vdev->memory_lock);
 	xa_init(&vdev->ctx);
 
@@ -2459,11 +2471,17 @@  static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set,
 	 * cause the PCI config space reset without restoring the original
 	 * state (saved locally in 'vdev->pm_save').
 	 */
-	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list)
+	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list) {
+		vfio_pci_dma_buf_move(vdev, true);
 		vfio_pci_set_power_state(vdev, PCI_D0);
+	}
 
 	ret = pci_reset_bus(pdev);
 
+	list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list)
+		if (__vfio_pci_memory_enabled(vdev))
+			vfio_pci_dma_buf_move(vdev, false);
+
 	vdev = list_last_entry(&dev_set->device_list,
 			       struct vfio_pci_core_device, vdev.dev_set_list);
 
diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h
index 5e4fa69aee16..09d3c300918c 100644
--- a/drivers/vfio/pci/vfio_pci_priv.h
+++ b/drivers/vfio/pci/vfio_pci_priv.h
@@ -101,4 +101,27 @@  static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
 	return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
 }
 
+#ifdef CONFIG_DMA_SHARED_BUFFER
+int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
+				  struct vfio_device_feature_dma_buf __user *arg,
+				  size_t argsz);
+void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev);
+void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked);
+#else
+static int
+vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
+			      struct vfio_device_feature_dma_buf __user *arg,
+			      size_t argsz)
+{
+	return -ENOTTY;
+}
+static inline void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
+{
+}
+static inline void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev,
+					 bool revoked)
+{
+}
+#endif
+
 #endif
diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h
index f87067438ed4..3756e0904de2 100644
--- a/include/linux/vfio_pci_core.h
+++ b/include/linux/vfio_pci_core.h
@@ -94,6 +94,7 @@  struct vfio_pci_core_device {
 	struct vfio_pci_core_device	*sriov_pf_core_dev;
 	struct notifier_block	nb;
 	struct rw_semaphore	memory_lock;
+	struct list_head	dmabufs;
 };
 
 /* Will be exported for vfio pci drivers usage */
diff --git a/include/uapi/linux/vfio.h b/include/uapi/linux/vfio.h
index 2b68e6cdf190..6f283dadacb7 100644
--- a/include/uapi/linux/vfio.h
+++ b/include/uapi/linux/vfio.h
@@ -1458,6 +1458,31 @@  struct vfio_device_feature_bus_master {
 };
 #define VFIO_DEVICE_FEATURE_BUS_MASTER 10
 
+/**
+ * Upon VFIO_DEVICE_FEATURE_GET create a dma_buf fd for the
+ * regions selected.
+ *
+ * open_flags are the typical flags passed to open(2), eg O_RDWR, O_CLOEXEC,
+ * etc. offset/length specify a slice of the region to create the dmabuf from.
+ * nr_ranges is the total number of (P2P DMA) ranges that comprise the dmabuf.
+ *
+ * Return: The fd number on success, -1 and errno is set on failure.
+ */
+#define VFIO_DEVICE_FEATURE_DMA_BUF 11
+
+struct vfio_region_dma_range {
+	__u32	region_index;
+	__u32	__pad;
+	__u64	offset;
+	__u64	length;
+};
+
+struct vfio_device_feature_dma_buf {
+	__u32	open_flags;
+	__u32	nr_ranges;
+	struct vfio_region_dma_range dma_ranges[];
+};
+
 /* -------- API for Type1 VFIO IOMMU -------- */
 
 /**