Re: [PATCH] dma-buf/poll: Get a file reference for outstanding fence callbacks

[Daniel Vetter <daniel@ffwll.ch>]

On Fri, Jul 23, 2021 at 10:19:49AM +0200, Michel Dänzer wrote:
> On 2021-07-23 10:04 a.m., Christian König wrote:
> > Am 23.07.21 um 09:58 schrieb Michel Dänzer:
> >> From: Michel Dänzer <mdaenzer@redhat.com>
> >>
> >> This makes sure we don't hit the
> >>
> >>     BUG_ON(dmabuf->cb_in.active || dmabuf->cb_out.active);
> >>
> >> in dma_buf_release, which could be triggered by user space closing the
> >> dma-buf file description while there are outstanding fence callbacks
> >> from dma_buf_poll.
> > 
> > I was also wondering the same thing while working on this, but then thought that the poll interface would take care of this.
> 
> I was able to hit the BUG_ON with https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1880 .

igt test would be really lovely. Maybe base something off the
import/export igts from Jason?
-Daniel

> 
> 
> >> Cc: stable@vger.kernel.org
> >> Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
> >> ---
> >>   drivers/dma-buf/dma-buf.c | 18 ++++++++++++------
> >>   1 file changed, 12 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c
> >> index 6c520c9bd93c..ec25498a971f 100644
> >> --- a/drivers/dma-buf/dma-buf.c
> >> +++ b/drivers/dma-buf/dma-buf.c
> >> @@ -65,12 +65,9 @@ static void dma_buf_release(struct dentry *dentry)
> >>       BUG_ON(dmabuf->vmapping_counter);
> >>         /*
> >> -     * Any fences that a dma-buf poll can wait on should be signaled
> >> -     * before releasing dma-buf. This is the responsibility of each
> >> -     * driver that uses the reservation objects.
> >> -     *
> >> -     * If you hit this BUG() it means someone dropped their ref to the
> >> -     * dma-buf while still having pending operation to the buffer.
> >> +     * If you hit this BUG() it could mean:
> >> +     * * There's a file reference imbalance in dma_buf_poll / dma_buf_poll_cb or somewhere else
> >> +     * * dmabuf->cb_in/out.active are non-0 despite no pending fence callback
> >>        */
> >>       BUG_ON(dmabuf->cb_in.active || dmabuf->cb_out.active);
> >>   @@ -196,6 +193,7 @@ static loff_t dma_buf_llseek(struct file *file, loff_t offset, int whence)
> >>   static void dma_buf_poll_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
> >>   {
> >>       struct dma_buf_poll_cb_t *dcb = (struct dma_buf_poll_cb_t *)cb;
> >> +    struct dma_buf *dmabuf = container_of(dcb->poll, struct dma_buf, poll);
> >>       unsigned long flags;
> >>         spin_lock_irqsave(&dcb->poll->lock, flags);
> >> @@ -203,6 +201,8 @@ static void dma_buf_poll_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
> >>       dcb->active = 0;
> >>       spin_unlock_irqrestore(&dcb->poll->lock, flags);
> >>       dma_fence_put(fence);
> >> +    /* Paired with get_file in dma_buf_poll */
> >> +    fput(dmabuf->file);
> > 
> > Is calling fput() in interrupt context ok? IIRC that could potentially sleep.
> 
> Looks fine AFAICT: It has
> 
> 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
> 
> and as a fallback for that, it adds the file to a lock-less delayed_fput_list which is processed by a workqueue.
> 
> 
> -- 
> Earthling Michel Dänzer               |               https://redhat.com
> Libre software enthusiast             |             Mesa and X developer

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch