From: Leon Romanovsky <leon@kernel.org>
To: "Christian König" <christian.koenig@amd.com>
Cc: "Sumit Semwal" <sumit.semwal@linaro.org>,
"Alex Deucher" <alexander.deucher@amd.com>,
"David Airlie" <airlied@gmail.com>,
"Simona Vetter" <simona@ffwll.ch>,
"Gerd Hoffmann" <kraxel@redhat.com>,
"Dmitry Osipenko" <dmitry.osipenko@collabora.com>,
"Gurchetan Singh" <gurchetansingh@chromium.org>,
"Chia-I Wu" <olvaffe@gmail.com>,
"Maarten Lankhorst" <maarten.lankhorst@linux.intel.com>,
"Maxime Ripard" <mripard@kernel.org>,
"Thomas Zimmermann" <tzimmermann@suse.de>,
"Lucas De Marchi" <lucas.demarchi@intel.com>,
"Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
"Rodrigo Vivi" <rodrigo.vivi@intel.com>,
"Jason Gunthorpe" <jgg@ziepe.ca>,
"Kevin Tian" <kevin.tian@intel.com>,
"Joerg Roedel" <joro@8bytes.org>, "Will Deacon" <will@kernel.org>,
"Robin Murphy" <robin.murphy@arm.com>,
"Felix Kuehling" <Felix.Kuehling@amd.com>,
"Alex Williamson" <alex@shazbot.org>,
"Ankit Agrawal" <ankita@nvidia.com>,
"Vivek Kasireddy" <vivek.kasireddy@intel.com>,
linux-media@vger.kernel.org, dri-devel@lists.freedesktop.org,
linaro-mm-sig@lists.linaro.org, linux-kernel@vger.kernel.org,
amd-gfx@lists.freedesktop.org, virtualization@lists.linux.dev,
intel-xe@lists.freedesktop.org, linux-rdma@vger.kernel.org,
iommu@lists.linux.dev, kvm@vger.kernel.org
Subject: Re: [PATCH v5 4/8] vfio: Wait for dma-buf invalidation to complete
Date: Fri, 30 Jan 2026 15:01:31 +0200 [thread overview]
Message-ID: <20260130130131.GO10992@unreal> (raw)
In-Reply-To: <31872c87-5cba-4081-8196-72cc839c6122@amd.com>
On Fri, Jan 30, 2026 at 09:30:59AM +0100, Christian König wrote:
> On 1/24/26 20:14, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro@nvidia.com>
> >
> > dma-buf invalidation is handled asynchronously by the hardware, so VFIO
> > must wait until all affected objects have been fully invalidated.
> >
> > In addition, the dma-buf exporter is expecting that all importers unmap any
> > buffers they previously mapped.
> >
> > Fixes: 5d74781ebc86 ("vfio/pci: Add dma-buf export support for MMIO regions")
> > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > ---
> > drivers/vfio/pci/vfio_pci_dmabuf.c | 71 ++++++++++++++++++++++++++++++++++++--
> > 1 file changed, 68 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c
> > index d8ceafabef48..485515629fe4 100644
> > --- a/drivers/vfio/pci/vfio_pci_dmabuf.c
> > +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c
> > @@ -17,6 +17,8 @@ struct vfio_pci_dma_buf {
> > struct dma_buf_phys_vec *phys_vec;
> > struct p2pdma_provider *provider;
> > u32 nr_ranges;
> > + struct kref kref;
> > + struct completion comp;
> > u8 revoked : 1;
> > };
> >
> > @@ -44,27 +46,46 @@ static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
> > return 0;
> > }
> >
> > +static void vfio_pci_dma_buf_done(struct kref *kref)
> > +{
> > + struct vfio_pci_dma_buf *priv =
> > + container_of(kref, struct vfio_pci_dma_buf, kref);
> > +
> > + complete(&priv->comp);
> > +}
> > +
> > static struct sg_table *
> > vfio_pci_dma_buf_map(struct dma_buf_attachment *attachment,
> > enum dma_data_direction dir)
> > {
> > struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> > + struct sg_table *ret;
> >
> > dma_resv_assert_held(priv->dmabuf->resv);
> >
> > if (priv->revoked)
> > return ERR_PTR(-ENODEV);
> >
> > - return dma_buf_phys_vec_to_sgt(attachment, priv->provider,
> > - priv->phys_vec, priv->nr_ranges,
> > - priv->size, dir);
> > + ret = dma_buf_phys_vec_to_sgt(attachment, priv->provider,
> > + priv->phys_vec, priv->nr_ranges,
> > + priv->size, dir);
> > + if (IS_ERR(ret))
> > + return ret;
> > +
> > + kref_get(&priv->kref);
> > + return 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;
> > +
> > + dma_resv_assert_held(priv->dmabuf->resv);
> > +
> > dma_buf_free_sgt(attachment, sgt, dir);
> > + kref_put(&priv->kref, vfio_pci_dma_buf_done);
> > }
> >
> > static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf)
> > @@ -287,6 +308,9 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
> > goto err_dev_put;
> > }
> >
> > + kref_init(&priv->kref);
> > + init_completion(&priv->comp);
> > +
> > /* dma_buf_put() now frees priv */
> > INIT_LIST_HEAD(&priv->dmabufs_elm);
> > down_write(&vdev->memory_lock);
> > @@ -326,6 +350,8 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
> > lockdep_assert_held_write(&vdev->memory_lock);
> >
> > list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
> > + unsigned long wait;
> > +
> > if (!get_file_active(&priv->dmabuf->file))
> > continue;
> >
> > @@ -333,7 +359,37 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
> > dma_resv_lock(priv->dmabuf->resv, NULL);
> > priv->revoked = revoked;
> > dma_buf_invalidate_mappings(priv->dmabuf);
> > + dma_resv_wait_timeout(priv->dmabuf->resv,
> > + DMA_RESV_USAGE_BOOKKEEP, false,
> > + MAX_SCHEDULE_TIMEOUT);
> > dma_resv_unlock(priv->dmabuf->resv);
> > + if (revoked) {
> > + kref_put(&priv->kref, vfio_pci_dma_buf_done);
> > + /* Let's wait till all DMA unmap are completed. */
> > + wait = wait_for_completion_timeout(
> > + &priv->comp, secs_to_jiffies(1));
> > + /*
> > + * If you see this WARN_ON, it means that
> > + * importer didn't call unmap in response to
> > + * dma_buf_invalidate_mappings() which is not
> > + * allowed.
> > + */
> > + WARN(!wait,
> > + "Timed out waiting for DMABUF unmap, importer has a broken invalidate_mapping()");
>
> You can do the revoke to do your resource management, for example re-use the backing store for something else.
>
> But it is mandatory that you keep the mapping around indefinitely until the importer closes it.
>
> Before that you can't do things like runtime PM or remove or anything which would make the DMA addresses invalid.
>
> As far as I can see vfio_pci_dma_buf_move() is used exactly for that use case so this here is an absolutely clear NAK from my side for this approach.
>
> You can either split up the functionality of vfio_pci_dma_buf_move() into vfio_pci_dma_buf_invalidate_mappings() and vfio_pci_dma_buf_flush() and then call the later whenever necessary or you keep it in one function and block everybody until the importer has dropped all mappings.
No problem, I can change it to be:
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c
index d087d018d547..53772a84c93b 100644
--- a/drivers/vfio/pci/vfio_pci_dmabuf.c
+++ b/drivers/vfio/pci/vfio_pci_dmabuf.c
@@ -357,23 +357,7 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
dma_resv_unlock(priv->dmabuf->resv);
if (revoked) {
kref_put(&priv->kref, vfio_pci_dma_buf_done);
- /*
- * Let's wait for 1 second till all DMA unmap
- * are completed. It is supposed to catch dma-buf
- * importers which lied about their support
- * of dmabuf revoke. See dma_buf_invalidate_mappings()
- * for the expected behaviour.
- */
- wait = wait_for_completion_timeout(
- &priv->comp, secs_to_jiffies(1));
- /*
- * If you see this WARN_ON, it means that
- * importer didn't call unmap in response to
- * dma_buf_invalidate_mappings() which is not
- * allowed.
- */
- WARN(!wait,
- "Timed out waiting for DMABUF unmap, importer has a broken invalidate_mapping()");
+ wait_for_completion(&priv->comp);
} else {
/*
* Kref is initialize again, because when revoke
Do you want me to send v6?
Thanks
>
> > + } else {
> > + /*
> > + * Kref is initialize again, because when revoke
> > + * was performed the reference counter was decreased
> > + * to zero to trigger completion.
> > + */
> > + kref_init(&priv->kref);
> > + /*
> > + * There is no need to wait as no mapping was
> > + * performed when the previous status was
> > + * priv->revoked == true.
> > + */
> > + reinit_completion(&priv->comp);
> > + }
> > }
> > fput(priv->dmabuf->file);
>
> This is also extremely questionable. Why doesn't the dmabuf have a reference while on the linked list?
>
> Regards,
> Christian.
>
> > }
> > @@ -346,6 +402,8 @@ void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
> >
> > down_write(&vdev->memory_lock);
> > list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
> > + unsigned long wait;
> > +
> > if (!get_file_active(&priv->dmabuf->file))
> > continue;
> >
> > @@ -354,7 +412,14 @@ void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev)
> > priv->vdev = NULL;
> > priv->revoked = true;
> > dma_buf_invalidate_mappings(priv->dmabuf);
> > + dma_resv_wait_timeout(priv->dmabuf->resv,
> > + DMA_RESV_USAGE_BOOKKEEP, false,
> > + MAX_SCHEDULE_TIMEOUT);
> > dma_resv_unlock(priv->dmabuf->resv);
> > + kref_put(&priv->kref, vfio_pci_dma_buf_done);
> > + wait = wait_for_completion_timeout(&priv->comp,
> > + secs_to_jiffies(1));
> > + WARN_ON(!wait);
> > vfio_device_put_registration(&vdev->vdev);
> > fput(priv->dmabuf->file);
> > }
> >
>
>
next prev parent reply other threads:[~2026-01-30 13:01 UTC|newest]
Thread overview: 51+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-01-24 19:14 [PATCH v5 0/8] dma-buf: Use revoke mechanism to invalidate shared buffers Leon Romanovsky
2026-01-24 19:14 ` [PATCH v5 1/8] dma-buf: Rename .move_notify() callback to a clearer identifier Leon Romanovsky
2026-01-24 19:14 ` [PATCH v5 2/8] dma-buf: Rename dma_buf_move_notify() to dma_buf_invalidate_mappings() Leon Romanovsky
2026-01-27 9:21 ` Christian König
2026-01-24 19:14 ` [PATCH v5 3/8] dma-buf: Always build with DMABUF_MOVE_NOTIFY Leon Romanovsky
2026-01-27 9:26 ` Christian König
2026-01-27 9:58 ` Leon Romanovsky
2026-01-27 10:02 ` Christian König
2026-01-27 11:42 ` Leon Romanovsky
2026-01-27 20:45 ` Leon Romanovsky
2026-01-30 7:00 ` Leon Romanovsky
2026-01-24 19:14 ` [PATCH v5 4/8] vfio: Wait for dma-buf invalidation to complete Leon Romanovsky
2026-01-26 20:53 ` Pranjal Shrivastava
2026-01-27 8:58 ` Leon Romanovsky
2026-01-27 16:27 ` Jason Gunthorpe
2026-01-29 7:06 ` Tian, Kevin
2026-01-29 7:33 ` Leon Romanovsky
2026-01-29 8:13 ` Tian, Kevin
2026-01-29 8:41 ` Leon Romanovsky
2026-01-29 21:04 ` Alex Williamson
2026-01-30 3:10 ` Tian, Kevin
2026-01-29 14:58 ` Jason Gunthorpe
2026-01-30 3:12 ` Tian, Kevin
2026-01-30 5:43 ` Mauro Carvalho Chehab
2026-01-30 5:48 ` Tian, Kevin
2026-01-30 8:46 ` Christian König
2026-01-30 8:30 ` Christian König
2026-01-30 13:01 ` Leon Romanovsky [this message]
2026-01-30 13:21 ` Christian König
2026-01-30 13:31 ` Leon Romanovsky
2026-01-30 13:56 ` Jason Gunthorpe
2026-01-30 14:11 ` Christian König
2026-01-30 14:44 ` Jason Gunthorpe
2026-02-02 8:42 ` Christian König
2026-02-02 15:12 ` Jason Gunthorpe
2026-02-02 15:21 ` Christian König
2026-02-02 15:55 ` Jason Gunthorpe
2026-01-24 19:14 ` [PATCH v5 5/8] dma-buf: Make .invalidate_mapping() truly optional Leon Romanovsky
2026-01-30 8:30 ` Christian König
2026-01-30 12:55 ` Leon Romanovsky
2026-01-24 19:14 ` [PATCH v5 6/8] dma-buf: Add dma_buf_attach_revocable() Leon Romanovsky
2026-01-26 20:38 ` Pranjal Shrivastava
2026-01-26 21:01 ` Jason Gunthorpe
2026-01-30 8:43 ` Christian König
2026-01-30 14:00 ` Jason Gunthorpe
2026-01-24 19:14 ` [PATCH v5 7/8] vfio: Permit VFIO to work with pinned importers Leon Romanovsky
2026-01-29 21:04 ` Alex Williamson
2026-01-30 3:14 ` Tian, Kevin
2026-01-24 19:14 ` [PATCH v5 8/8] iommufd: Add dma_buf_pin() Leon Romanovsky
2026-01-29 7:08 ` Tian, Kevin
2026-01-30 0:17 ` Jason Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260130130131.GO10992@unreal \
--to=leon@kernel.org \
--cc=Felix.Kuehling@amd.com \
--cc=airlied@gmail.com \
--cc=alex@shazbot.org \
--cc=alexander.deucher@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=ankita@nvidia.com \
--cc=christian.koenig@amd.com \
--cc=dmitry.osipenko@collabora.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=gurchetansingh@chromium.org \
--cc=intel-xe@lists.freedesktop.org \
--cc=iommu@lists.linux.dev \
--cc=jgg@ziepe.ca \
--cc=joro@8bytes.org \
--cc=kevin.tian@intel.com \
--cc=kraxel@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=linaro-mm-sig@lists.linaro.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=lucas.demarchi@intel.com \
--cc=maarten.lankhorst@linux.intel.com \
--cc=mripard@kernel.org \
--cc=olvaffe@gmail.com \
--cc=robin.murphy@arm.com \
--cc=rodrigo.vivi@intel.com \
--cc=simona@ffwll.ch \
--cc=sumit.semwal@linaro.org \
--cc=thomas.hellstrom@linux.intel.com \
--cc=tzimmermann@suse.de \
--cc=virtualization@lists.linux.dev \
--cc=vivek.kasireddy@intel.com \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox