Re: [PATCH v5 2/9] ALSA: virtio: add virtio sound driver

[Anton Yakovlev]

On 25.02.2021 11:38, Takashi Iwai wrote:
> On Mon, 22 Feb 2021 16:34:37 +0100,
> Anton Yakovlev wrote:
> > +static int virtsnd_find_vqs(struct virtio_snd *snd)
> > +{
> > +     struct virtio_device *vdev = snd->vdev;
> > +     vq_callback_t *callbacks[VIRTIO_SND_VQ_MAX] = {
> > +             [VIRTIO_SND_VQ_EVENT] = virtsnd_event_notify_cb
> > +     };
> > +     const char *names[VIRTIO_SND_VQ_MAX] = {
>
> Shouldn't be static?

Well, yes. Although in this particular case, I do not think it is that
critical. :)


> Also it's often const char * const names[] = { ... }
> unless you overwrite something.

I tried to use the same type names as in the function prototype.
Otherwise the compiler or static analyzer may complain.


> > +/**
> > + * virtsnd_reset_fn() - Kernel worker's function to reset the device.
> > + * @work: Reset device work.
> > + *
> > + * Context: Process context.
> > + */
> > +static void virtsnd_reset_fn(struct work_struct *work)
> > +{
> > +     struct virtio_snd *snd =
> > +             container_of(work, struct virtio_snd, reset_work);
> > +     struct virtio_device *vdev = snd->vdev;
> > +     struct device *dev = &vdev->dev;
> > +     int rc;
> > +
> > +     dev_info(dev, "sound device needs reset\n");
> > +
> > +     /*
> > +      * It seems that the only way to properly reset the device is to remove
> > +      * and re-create the ALSA sound card device.
> > +      */
> > +     rc = device_reprobe(dev);
> > +     if (rc)
> > +             dev_err(dev, "failed to reprobe sound device: %d\n", rc);
>
> Now I'm wondering whether it's safe to do that from this place.
> Basically device_reprobe() unbinds the device that releases the full
> resources once including the devm_* stuff.  And this work itself is in
> a part of devm allocated resource, so it'll be released there.  That
> said, we might hit use-after-free...  This needs to be verified.

It's safe. Suicide kernel workers are funny but possible things. Since
the kernel itself (AFAIU) assumes such a situation and does not access
the worker structure after the callback function call.


> thanks,
>
> Takashi

--
Anton Yakovlev
Senior Software Engineer

OpenSynergy GmbH
Rotherstr. 20, 10245 Berlin