* [PATCH] vfio: pci: use kzalloc_flex
@ 2026-03-26 2:37 Rosen Penev
2026-03-30 22:19 ` Alex Williamson
0 siblings, 1 reply; 8+ messages in thread
From: Rosen Penev @ 2026-03-26 2:37 UTC (permalink / raw)
To: kvm
Cc: Alex Williamson, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b
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.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
---
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:
--
2.53.0
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-26 2:37 [PATCH] vfio: pci: use kzalloc_flex Rosen Penev
@ 2026-03-30 22:19 ` Alex Williamson
2026-03-30 23:24 ` Rosen Penev
0 siblings, 1 reply; 8+ messages in thread
From: Alex Williamson @ 2026-03-30 22:19 UTC (permalink / raw)
To: Rosen Penev
Cc: kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
alex, Leon Romanovsky
[Cc +Leon]
On Wed, 25 Mar 2026 19:37:47 -0700
Rosen Penev <rosenp@gmail.com> 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.
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.
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@gmail.com>
> ---
> 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:
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-30 22:19 ` Alex Williamson
@ 2026-03-30 23:24 ` Rosen Penev
2026-03-30 23:46 ` Gustavo A. R. Silva
2026-03-31 0:23 ` Alex Williamson
0 siblings, 2 replies; 8+ messages in thread
From: Rosen Penev @ 2026-03-30 23:24 UTC (permalink / raw)
To: Alex Williamson
Cc: kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
Leon Romanovsky
On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@shazbot.org> wrote:
>
> [Cc +Leon]
>
> On Wed, 25 Mar 2026 19:37:47 -0700
> Rosen Penev <rosenp@gmail.com> 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.
>
> 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@gmail.com>
> > ---
> > 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:
>
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-30 23:24 ` Rosen Penev
@ 2026-03-30 23:46 ` Gustavo A. R. Silva
2026-03-31 0:18 ` Rosen Penev
2026-03-31 0:23 ` Alex Williamson
1 sibling, 1 reply; 8+ messages in thread
From: Gustavo A. R. Silva @ 2026-03-30 23:46 UTC (permalink / raw)
To: Rosen Penev, Alex Williamson
Cc: kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
Leon Romanovsky
On 3/30/26 17:24, Rosen Penev wrote:
> On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@shazbot.org> wrote:
>>
>> [Cc +Leon]
>>
>> On Wed, 25 Mar 2026 19:37:47 -0700
>> Rosen Penev <rosenp@gmail.com> 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.
Not true. However, it's best practice.
The requirement is that the _counter_ must be initialized before
the first reference to the flexible-array member.
-Gustavo
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-30 23:46 ` Gustavo A. R. Silva
@ 2026-03-31 0:18 ` Rosen Penev
2026-03-31 0:51 ` Gustavo A. R. Silva
0 siblings, 1 reply; 8+ messages in thread
From: Rosen Penev @ 2026-03-31 0:18 UTC (permalink / raw)
To: Gustavo A. R. Silva
Cc: Alex Williamson, kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
Leon Romanovsky
On Mon, Mar 30, 2026 at 4:47 PM Gustavo A. R. Silva
<gustavo@embeddedor.com> wrote:
>
>
>
> On 3/30/26 17:24, Rosen Penev wrote:
> > On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@shazbot.org> wrote:
> >>
> >> [Cc +Leon]
> >>
> >> On Wed, 25 Mar 2026 19:37:47 -0700
> >> Rosen Penev <rosenp@gmail.com> 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.
>
> Not true. However, it's best practice.
>
> The requirement is that the _counter_ must be initialized before
> the first reference to the flexible-array member.
OTOH kzalloc_flex automatically sets it for GCC15 and above. Useful to
keep it right after for an eventual coccinelle script...
>
> -Gustavo
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-30 23:24 ` Rosen Penev
2026-03-30 23:46 ` Gustavo A. R. Silva
@ 2026-03-31 0:23 ` Alex Williamson
2026-03-31 5:11 ` Leon Romanovsky
1 sibling, 1 reply; 8+ messages in thread
From: Alex Williamson @ 2026-03-31 0:23 UTC (permalink / raw)
To: Rosen Penev
Cc: kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
Leon Romanovsky
On Mon, Mar 30, 2026, at 5:24 PM, Rosen Penev wrote:
> On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@shazbot.org> wrote:
>>
>> [Cc +Leon]
>>
>> On Wed, 25 Mar 2026 19:37:47 -0700
>> Rosen Penev <rosenp@gmail.com> 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@gmail.com>
>> > ---
>> > 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:
>>
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-31 0:18 ` Rosen Penev
@ 2026-03-31 0:51 ` Gustavo A. R. Silva
0 siblings, 0 replies; 8+ messages in thread
From: Gustavo A. R. Silva @ 2026-03-31 0:51 UTC (permalink / raw)
To: Rosen Penev
Cc: Alex Williamson, kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b,
Leon Romanovsky
>>>>> 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.
>>
>> Not true. However, it's best practice.
>>
>> The requirement is that the _counter_ must be initialized before
>> the first reference to the flexible-array member.
> OTOH kzalloc_flex automatically sets it for GCC15 and above. Useful to
That's what the "it's best practice." comment above alludes to.
-Gustavo
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] vfio: pci: use kzalloc_flex
2026-03-31 0:23 ` Alex Williamson
@ 2026-03-31 5:11 ` Leon Romanovsky
0 siblings, 0 replies; 8+ messages in thread
From: Leon Romanovsky @ 2026-03-31 5:11 UTC (permalink / raw)
To: Alex Williamson
Cc: Rosen Penev, kvm, Kees Cook, Gustavo A. R. Silva, open list,
open list:KERNEL HARDENING (not covered by other areas):Keyword:b__counted_by(_le|_be)?b
On Mon, Mar 30, 2026 at 06:23:05PM -0600, Alex Williamson wrote:
> On Mon, Mar 30, 2026, at 5:24 PM, Rosen Penev wrote:
> > On Mon, Mar 30, 2026 at 4:16 PM Alex Williamson <alex@shazbot.org> wrote:
> >>
> >> [Cc +Leon]
> >>
> >> On Wed, 25 Mar 2026 19:37:47 -0700
> >> Rosen Penev <rosenp@gmail.com> 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,
My comment is that we can simply drop this patch. It adds no value and only
creates unnecessary churn.
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@gmail.com>
> >> > ---
> >> > 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:
> >>
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2026-03-31 5:11 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-03-26 2:37 [PATCH] vfio: pci: use kzalloc_flex Rosen Penev
2026-03-30 22:19 ` Alex Williamson
2026-03-30 23:24 ` Rosen Penev
2026-03-30 23:46 ` Gustavo A. R. Silva
2026-03-31 0:18 ` Rosen Penev
2026-03-31 0:51 ` Gustavo A. R. Silva
2026-03-31 0:23 ` Alex Williamson
2026-03-31 5:11 ` Leon Romanovsky
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox