| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20260128122439-mutt-send-email-mst@kernel.org>
Date: Wed, 28 Jan 2026 12:30:03 -0500
From: "Michael S. Tsirkin" <mst@...hat.com>
To: Alexander Graf <graf@...zon.com>
Cc: Johannes Thumshirn <johannes.thumshirn@....com>,
Jason Wang <jasowang@...hat.com>,
Xuan Zhuo <xuanzhuo@...ux.alibaba.com>,
Eugenio Pérez <eperezma@...hat.com>,
"open list:VIRTIO CORE" <virtualization@...ts.linux.dev>,
open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] virtio_ring: Add READ_ONCE annotations for
device-writable fields
On Wed, Jan 28, 2026 at 03:47:27PM +0100, Alexander Graf wrote:
>
> On 28.01.26 14:59, Johannes Thumshirn wrote:
> > From: Alexander Graf <graf@...zon.com>
> >
> > KCSAN reports data races when accessing virtio ring fields that are
> > concurrently written by the device (host). These are legitimate
> > concurrent accesses where the CPU reads fields that the device updates
> > via DMA-like mechanisms.
> >
> > Add accessor functions that use READ_ONCE() to properly annotate these
> > device-writable fields and prevent compiler optimizations that could
> > break the code. This also serves as documentation showing which fields
> > are shared with the device.
> >
> > The affected fields are:
> > - Split ring: used->idx, used->ring[].id, used->ring[].len
> > - Packed ring: desc[].flags, desc[].id, desc[].len
> >
> > Reported-by: Kernel Concurrency Sanitizer (KCSAN)
> > Signed-off-by: Alexander Graf <graf@...zon.com>
>
>
> Thanks for persistently trying to fix these KCSAN warnings! :)
>
> This patch was an initial AI generated stab at seeing whether READ_ONCE
> would work and how to make it pretty. It was not meant to go to the mailing
> list as is. Some comments on what we would need to improve to bring it to a
> mergeable state.
According to latest Docs, use of AI should be documented.
Just a statement to this end is probably going to be enough.
>
> Given this is not a subsystem-contributor relationship, I also think it
> would be Co-developed-by instead of signed-off-by :).
>
> > [jth: Add READ_ONCE in virtqueue_kick_prepare_split ]
> > Signed-off-by: Johannes Thumshirn <johannes.thumshirn@....com>
> > ---
> > drivers/virtio/virtio_ring.c | 88 ++++++++++++++++++++++++++++++------
> > 1 file changed, 73 insertions(+), 15 deletions(-)
> >
> > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> > index ddab68959671..74957c83e138 100644
> > --- a/drivers/virtio/virtio_ring.c
> > +++ b/drivers/virtio/virtio_ring.c
> > @@ -222,6 +222,63 @@ struct vring_virtqueue {
> > #endif
> > };
> >
> > +/*
> > + * Accessors for device-writable fields in virtio rings.
> > + * These fields are concurrently written by the device and read by the driver.
> > + * Use READ_ONCE() to prevent compiler optimizations and document the
> > + * intentional data race.
>
>
> Should mention that this is necessary for KCSAN.
>
>
> > + */
> > +
> > +/* Split ring: read device-written fields from used ring */
>
>
> Useless comment
>
>
> > +static inline u16 vring_used_idx_read(const struct vring_virtqueue *vq)
>
>
> Just do a complete sed s/_read// on this patch. Nobody needs these _read
> suffixes.
That's fine, too.
>
> > +{
> > + return virtio16_to_cpu(vq->vq.vdev,
> > + READ_ONCE(vq->split.vring.used->idx));
> > +}
> > +
> > +static inline u32 vring_used_id_read(const struct vring_virtqueue *vq,
> > + u16 idx)
> > +{
> > + return virtio32_to_cpu(vq->vq.vdev,
> > + READ_ONCE(vq->split.vring.used->ring[idx].id));
> > +}
> > +
> > +static inline u32 vring_used_len_read(const struct vring_virtqueue *vq,
> > + u16 idx)
> > +{
> > + return virtio32_to_cpu(vq->vq.vdev,
> > + READ_ONCE(vq->split.vring.used->ring[idx].len));
> > +}
> > +
> > +/* Packed ring: read device-written fields from descriptors */
>
>
> Useless comment
>
>
> > +static inline u16 vring_packed_desc_flags_read(const struct vring_virtqueue *vq,
> > + u16 idx)
> > +{
> > + return le16_to_cpu(READ_ONCE(vq->packed.vring.desc[idx].flags));
> > +}
> > +
> > +static inline u16 vring_packed_desc_id_read(const struct vring_virtqueue *vq,
> > + u16 idx)
> > +{
> > + return le16_to_cpu(READ_ONCE(vq->packed.vring.desc[idx].id));
> > +}
> > +
> > +static inline u32 vring_packed_desc_len_read(const struct vring_virtqueue *vq,
> > + u16 idx)
> > +{
> > + return le32_to_cpu(READ_ONCE(vq->packed.vring.desc[idx].len));
> > +}
> > +
> > +/*
> > + * Note: We don't need READ_ONCE for driver->device fields like:
> > + * - split.vring.avail->idx (driver writes, device reads)
> > + * - packed.vring.desc[].addr (driver writes, device reads)
> > + * These are written by the driver and only read by the device, so the
> > + * driver can safely access them without READ_ONCE. The device must use
> > + * appropriate barriers on its side.
> > + */
>
>
> Useless comment really. If you think it's worthwhile to mention the above,
> put it into the patch description.
>
>
> > +
> > +
> > static struct vring_desc_extra *vring_alloc_desc_extra(unsigned int num);
> > static void vring_free(struct virtqueue *_vq);
> >
> > @@ -736,9 +793,10 @@ static bool virtqueue_kick_prepare_split(struct virtqueue *_vq)
> > LAST_ADD_TIME_INVALID(vq);
> >
> > if (vq->event) {
> > - needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev,
> > - vring_avail_event(&vq->split.vring)),
> > - new, old);
> > + u16 event = virtio16_to_cpu(_vq->vdev,
> > + READ_ONCE(vring_avail_event(&vq->split.vring)));
> > +
> > + needs_kick = vring_need_event(event, new, old);
> > } else {
> > needs_kick = !(vq->split.vring.used->flags &
> > cpu_to_virtio16(_vq->vdev,
> > @@ -808,8 +866,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head,
> >
> > static bool more_used_split(const struct vring_virtqueue *vq)
> > {
> > - return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev,
> > - vq->split.vring.used->idx);
> > + return vq->last_used_idx != vring_used_idx_read(vq);
> > }
> >
> > static void *virtqueue_get_buf_ctx_split(struct virtqueue *_vq,
> > @@ -838,10 +895,8 @@ static void *virtqueue_get_buf_ctx_split(struct virtqueue *_vq,
> > virtio_rmb(vq->weak_barriers);
> >
> > last_used = (vq->last_used_idx & (vq->split.vring.num - 1));
> > - i = virtio32_to_cpu(_vq->vdev,
> > - vq->split.vring.used->ring[last_used].id);
> > - *len = virtio32_to_cpu(_vq->vdev,
> > - vq->split.vring.used->ring[last_used].len);
> > + i = vring_used_id_read(vq, last_used);
> > + *len = vring_used_len_read(vq, last_used);
> >
> > if (unlikely(i >= vq->split.vring.num)) {
> > BAD_RING(vq, "id %u out of range\n", i);
> > @@ -923,8 +978,7 @@ static bool virtqueue_poll_split(struct virtqueue *_vq, unsigned int last_used_i
> > {
> > struct vring_virtqueue *vq = to_vvq(_vq);
> >
> > - return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev,
> > - vq->split.vring.used->idx);
> > + return (u16)last_used_idx != vring_used_idx_read(vq);
> > }
> >
> > static bool virtqueue_enable_cb_delayed_split(struct virtqueue *_vq)
> > @@ -1701,10 +1755,10 @@ static void detach_buf_packed(struct vring_virtqueue *vq,
> > static inline bool is_used_desc_packed(const struct vring_virtqueue *vq,
> > u16 idx, bool used_wrap_counter)
> > {
> > - bool avail, used;
> > u16 flags;
> > + bool avail, used;
> >
> > - flags = le16_to_cpu(vq->packed.vring.desc[idx].flags);
> > + flags = vring_packed_desc_flags_read(vq, idx);
> > avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
> > used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
> >
> > @@ -1751,8 +1805,8 @@ static void *virtqueue_get_buf_ctx_packed(struct virtqueue *_vq,
> > last_used_idx = READ_ONCE(vq->last_used_idx);
> > used_wrap_counter = packed_used_wrap_counter(last_used_idx);
> > last_used = packed_last_used(last_used_idx);
> > - id = le16_to_cpu(vq->packed.vring.desc[last_used].id);
> > - *len = le32_to_cpu(vq->packed.vring.desc[last_used].len);
> > + id = vring_packed_desc_id_read(vq, last_used);
> > + *len = vring_packed_desc_len_read(vq, last_used);
> >
> > if (unlikely(id >= vq->packed.vring.num)) {
> > BAD_RING(vq, "id %u out of range\n", id);
> > @@ -1850,6 +1904,10 @@ static bool virtqueue_poll_packed(struct virtqueue *_vq, u16 off_wrap)
> > bool wrap_counter;
> > u16 used_idx;
> >
> > + /*
> > + * Note: off_wrap is from virtqueue_enable_cb_prepare_packed() which
> > + * already used READ_ONCE on vq->last_used_idx, so we don't need it again.
> > + */
>
>
> On its own in this code base 5 years from now, this comment will be super
> confusing. Because nobody has context what this note is about. I'd say just
> remove it.
>
>
> Alex
>
>
> > wrap_counter = off_wrap >> VRING_PACKED_EVENT_F_WRAP_CTR;
> > used_idx = off_wrap & ~(1 << VRING_PACKED_EVENT_F_WRAP_CTR);
> >
> > --
> > 2.52.0
> >
>
>
>
> Amazon Web Services Development Center Germany GmbH
> Tamara-Danz-Str. 13
> 10243 Berlin
> Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
> Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
> Sitz: Berlin
> Ust-ID: DE 365 538 597
Powered by blists - more mailing lists