[ofa-general] Re: [PATCH 2 of 9] Core of mmu notifiers

[Christoph Lameter]

On Tue, 8 Apr 2008, Andrea Arcangeli wrote:

> +	/*
> +	 * Called when nobody can register any more notifier in the mm
> +	 * and after the "mn" notifier has been disarmed already.
> +	 */
> +	void (*release)(struct mmu_notifier *mn,
> +			struct mm_struct *mm);

Hmmm... The unregister function does not call this. Guess driver calls
unregister function and does release like stuff on its own.

> +	/*
> +	 * invalidate_range_start() and invalidate_range_end() must be
> +	 * paired. Multiple invalidate_range_start/ends may be nested
> +	 * or called concurrently.
> +	 */

How could they be nested or called concurrently?


> +/*
> + * mm_users can't go down to zero while mmu_notifier_unregister()
> + * runs or it can race with ->release. So a mm_users pin must
> + * be taken by the caller (if mm can be different from current->mm).
> + */
> +int mmu_notifier_unregister(struct mmu_notifier *mn, struct mm_struct *mm)
> +{
> +	struct mm_lock_data *data;
> +
> +	BUG_ON(!atomic_read(&mm->mm_users));
> +
> +	data = mm_lock(mm);
> +	if (unlikely(IS_ERR(data)))
> +		return PTR_ERR(data);
> +	hlist_del(&mn->hlist);
> +	mm_unlock(mm, data);
> +	return 0;

Hmmm.. Ok, the user of the notifier does not get notified that it was 
unregistered.