Hi,
On Wed, 30 Aug 2023, Pierre-Louis Bossart wrote:
> > /* probe/remove/shutdown */
> > + int (*probe_no_wq)(struct snd_sof_dev *sof_dev); /* optional */
> > + int (*remove_no_wq)(struct snd_sof_dev *sof_dev); /* optional */
>
> My initial PR didn't have this remove_no_wq() callback.
>
> For symmetry it could be useful if it is meant to undo what the
> probe_no_wq() did, but conceptually the first thing we do in the remove
> is make sure that workqueue is either not scheduled or we wait until it
> completes.
I think that's exactly what the patch ends up with, remove_no_wq releases
resources acquired in probe_no_wq, i.e. they are symmetrical.
> int snd_sof_device_remove(struct device *dev)
> {
> struct snd_sof_dev *sdev = dev_get_drvdata(dev);
> struct snd_sof_pdata *pdata = sdev->pdata;
> int ret;
>
> if (IS_ENABLED(CONFIG_SND_SOC_SOF_PROBE_WORK_QUEUE))
> cancel_work_sync(&sdev->probe_work);
So this seems ok as well:
- if wq is used, at remove, we first wait wq to be finished
and only then proceed with removal
- the remove_no_wq is run only when the first step is completed
Br, Kai
