Now that I look at this more ...
> +static void cfg80211_wiphy_work(struct work_struct *work)
> +{
> + struct cfg80211_registered_device *rdev;
> +
> + rdev = container_of(work, struct cfg80211_registered_device, wiphy_work);
> +
> + spin_lock_irq(&rdev->wiphy_work_lock);
> + while (!list_empty(&rdev->wiphy_work_list)) {
> + struct wiphy_work *wk;
> +
> + wk = list_first_entry(&rdev->wiphy_work_list,
> + struct wiphy_work, entry);
> + list_del_init(&wk->entry);
> + spin_unlock_irq(&rdev->wiphy_work_lock);
> +
> + mutex_lock(&rdev->wiphy.mtx);
If I just change the locking here to take the wiphy.mtx before looking
at the list, which basically doesn't matter, then I don't even need
workqueue_pause() and all, nor do I even need a separate workqueue, just
schedule_work() will be good enough ... Just needs a _bit_ more work
when cancelling and here we should reschedule the work if the list isn't
empty after the first round, but overall that ends up far simpler.
So I think I'll drop the workqueue pause/resume the next time around,
FWIW.
johannes
