2023-06-28 08:56:43

by Tuo Li

Subject: [PATCH] media: bt8xx: Fix a possible data race in buffer_queue()

The variable btv->loop_irq is often protected by the lock btv->s_lock when
is accessed. Here is an example in bttv_irq_timeout():

spin_lock_irqsave(&btv->s_lock,flags);
...
btv->loop_irq = 0;
...
spin_unlock_irqrestore(&btv->s_lock,flags);

However, it is accessed without holding the lock btv->s_lock in
buffer_queue():

btv->loop_irq |= 1;

And thus a data race can occur.
To fix this possible data race, a lock and unlock pair is added when
accessing the variable btv->loop_irq in buffer_queue().

Reported-by: BassCheck <[email protected]>
Signed-off-by: Tuo Li <[email protected]>
---
drivers/media/pci/bt8xx/bttv-driver.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/drivers/media/pci/bt8xx/bttv-driver.c b/drivers/media/pci/bt8xx/bttv-driver.c
index 734f02b91aa3..241c51951627 100644
--- a/drivers/media/pci/bt8xx/bttv-driver.c
+++ b/drivers/media/pci/bt8xx/bttv-driver.c
@@ -1617,13 +1617,16 @@ buffer_queue(struct videobuf_queue *q, struct videobuf_buffer *vb)
struct bttv_buffer *buf = container_of(vb,struct bttv_buffer,vb);
struct bttv_fh *fh = q->priv_data;
struct bttv *btv = fh->btv;
+ unsigned long flags;

+ spin_lock_irqsave(&btv->s_lock,flags);
buf->vb.state = VIDEOBUF_QUEUED;
list_add_tail(&buf->vb.queue,&btv->capture);
if (!btv->curr.frame_irq) {
btv->loop_irq |= 1;
bttv_set_dma(btv, 0x03);
}
+ spin_unlock_irqrestore(&btv->s_lock,flags);
}

static void buffer_release(struct videobuf_queue *q, struct videobuf_buffer *vb)
--
2.34.1

2023-07-03 13:00:21

by Deborah Brouwer

Subject: Re: [PATCH] media: bt8xx: Fix a possible data race in buffer_queue()

On Wed, Jun 28, 2023 at 03:19:51PM +0800, Tuo Li wrote:
> The variable btv->loop_irq is often protected by the lock btv->s_lock when
> is accessed. Here is an example in bttv_irq_timeout():
>
> spin_lock_irqsave(&btv->s_lock,flags);
> ...
> btv->loop_irq = 0;
> ...
> spin_unlock_irqrestore(&btv->s_lock,flags);
>
> However, it is accessed without holding the lock btv->s_lock in
> buffer_queue():
>
> btv->loop_irq |= 1;
>
> And thus a data race can occur.
> To fix this possible data race, a lock and unlock pair is added when
> accessing the variable btv->loop_irq in buffer_queue().

hmm I am not sure why buffer_queue() didn't bother holding the
btv->s_lock but in the new version of the bttv driver that I have
been working on, the btv->s_lock is always held before changing
the loop_irq value.

https://patchwork.linuxtv.org/project/linux-media/cover/[email protected]/

When the bttv driver is converted to vb2, and the work is almost done,
the old buffer_queue() is totally eliminated, so I don't think it would
make sense to apply this patch now.

>
> Reported-by: BassCheck <[email protected]>
> Signed-off-by: Tuo Li <[email protected]>
> ---
> drivers/media/pci/bt8xx/bttv-driver.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/media/pci/bt8xx/bttv-driver.c b/drivers/media/pci/bt8xx/bttv-driver.c
> index 734f02b91aa3..241c51951627 100644
> --- a/drivers/media/pci/bt8xx/bttv-driver.c
> +++ b/drivers/media/pci/bt8xx/bttv-driver.c
> @@ -1617,13 +1617,16 @@ buffer_queue(struct videobuf_queue *q, struct videobuf_buffer *vb)
> struct bttv_buffer *buf = container_of(vb,struct bttv_buffer,vb);
> struct bttv_fh *fh = q->priv_data;
> struct bttv *btv = fh->btv;
> + unsigned long flags;
>
> + spin_lock_irqsave(&btv->s_lock,flags);
> buf->vb.state = VIDEOBUF_QUEUED;
> list_add_tail(&buf->vb.queue,&btv->capture);
> if (!btv->curr.frame_irq) {
> btv->loop_irq |= 1;
> bttv_set_dma(btv, 0x03);
> }
> + spin_unlock_irqrestore(&btv->s_lock,flags);
> }
>
> static void buffer_release(struct videobuf_queue *q, struct videobuf_buffer *vb)
> --
> 2.34.1
>