Re: [PATCH] vfio: pci: use kzalloc_flex

[Alex Williamson]

On Mon, Mar 30, 2026, at 5:24 PM, Rosen Penev wrote:
> On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@xxxxxxxxxxx> wrote:
>>
>> [Cc +Leon]
>>
>> On Wed, 25 Mar 2026 19:37:47 -0700
>> Rosen Penev <rosenp@xxxxxxxxx> wrote:
>>
>> > Simplify allocation by using a flexible array member and kzalloc_flex.
>> > Less memory management needed.
>> >
>> > Use __counted_by for extra runtime analysis. Move assignment to after
>> > allocation as required by __counted_by.
>>
>> I don't understand this statement, nr_ranges was previously set after
>> the allocation of phys_vec.  The only reordering was relative to
>> setting vdev, but that appears arbitrary.
> Yes that one. My understanding is __counted_by mandates immediate
> assignment after allocation. Otherwise UBSAN complains.
>>
>> In fact, we don't need to explicitly set the __counted_by variable at
>> all, kzalloc_flex() handles that.  So if anything, it's now redundant.
> Redundant with GCC`15 and above.

Sorry, this seems like a -2 on Rusty's manifesto of API design[1].  Why are we doing this?  We could do the math to do a single allocation and add the annotation for UBSAN without introducing this helper that's not obvious how to use correctly and at best introduces a redundant counter initialization.  Thanks,

Alex

[1] https://gist.github.com/mjball/9cd028ac793ae8b351df1379f1e721f9

>>
>> Leon, any other comments?  This should have a v2 removing the
>> redundancy and fixing the commit log.
>>
>> NB. This will be a bit messy to merge since kref and completion exist in
>> linux-next via drm, but maybe Linus will consolidate the hole in the
>> structure when he resolves it.  Thanks,
>>
>> Alex
>>
>> >
>> > Signed-off-by: Rosen Penev <rosenp@xxxxxxxxx>
>> > ---
>> >  drivers/vfio/pci/vfio_pci_dmabuf.c | 18 +++++-------------
>> >  1 file changed, 5 insertions(+), 13 deletions(-)
>> >
>> > diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c
>> > index 3a803923141b..40e7e035a720 100644
>> > --- a/drivers/vfio/pci/vfio_pci_dmabuf.c
>> > +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c
>> > @@ -14,12 +14,12 @@ struct vfio_pci_dma_buf {
>> >       struct vfio_pci_core_device *vdev;
>> >       struct list_head dmabufs_elm;
>> >       size_t size;
>> > -     struct phys_vec *phys_vec;
>> >       struct p2pdma_provider *provider;
>> >       u32 nr_ranges;
>> >       struct kref kref;
>> >       struct completion comp;
>> >       u8 revoked : 1;
>> > +     struct phys_vec phys_vec[] __counted_by(nr_ranges);
>> >  };
>> >
>> >  static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
>> > @@ -95,7 +95,6 @@ static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf)
>> >               up_write(&priv->vdev->memory_lock);
>> >               vfio_device_put_registration(&priv->vdev->vdev);
>> >       }
>> > -     kfree(priv->phys_vec);
>> >       kfree(priv);
>> >  }
>> >
>> > @@ -258,33 +257,28 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
>> >       if (ret)
>> >               goto err_free_ranges;
>> >
>> > -     priv = kzalloc_obj(*priv);
>> > +     priv = kzalloc_flex(*priv, phys_vec, get_dma_buf.nr_ranges);
>> >       if (!priv) {
>> >               ret = -ENOMEM;
>> >               goto err_free_ranges;
>> >       }
>> > -     priv->phys_vec = kzalloc_objs(*priv->phys_vec, get_dma_buf.nr_ranges);
>> > -     if (!priv->phys_vec) {
>> > -             ret = -ENOMEM;
>> > -             goto err_free_priv;
>> > -     }
>> >
>> > -     priv->vdev = vdev;
>> >       priv->nr_ranges = get_dma_buf.nr_ranges;
>> > +     priv->vdev = vdev;
>> >       priv->size = length;
>> >       ret = vdev->pci_ops->get_dmabuf_phys(vdev, &priv->provider,
>> >                                            get_dma_buf.region_index,
>> >                                            priv->phys_vec, dma_ranges,
>> >                                            priv->nr_ranges);
>> >       if (ret)
>> > -             goto err_free_phys;
>> > +             goto err_free_priv;
>> >
>> >       kfree(dma_ranges);
>> >       dma_ranges = NULL;
>> >
>> >       if (!vfio_device_try_get_registration(&vdev->vdev)) {
>> >               ret = -ENODEV;
>> > -             goto err_free_phys;
>> > +             goto err_free_priv;
>> >       }
>> >
>> >       exp_info.ops = &vfio_pci_dmabuf_ops;
>> > @@ -323,8 +317,6 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
>> >       dma_buf_put(priv->dmabuf);
>> >  err_dev_put:
>> >       vfio_device_put_registration(&vdev->vdev);
>> > -err_free_phys:
>> > -     kfree(priv->phys_vec);
>> >  err_free_priv:
>> >       kfree(priv);
>> >  err_free_ranges:
>>