On Mon, Mar 06, 2017 at 09:03:26PM -0800, Davidlohr Bueso wrote:
> +/*
> + * Implementation of read/write range locks.
> + *
> + * We keep interval tree of locked and to-be-locked ranges. When new range
> lock
> + * is requested, we add its interval to the tree and store number of
> intervals
> + * intersecting it to 'blocking_ranges'.
> + *
> + * When a range is unlocked, we again walk intervals that intersect with the
> + * unlocked one and decrement their 'blocking_ranges'. We wake up owner of
> any
> + * range lock whose 'blocking_ranges' drops to 0. For the shared case, the
> + * 'blocking_ranges' is only incremented if the intersecting range is not
> marked
> + * as a reader.
Not a word about fairness and starvation... Such important topics for
lock primitives.
In order to mitigate some of the tree walk overhead for
> + * non-intersecting ranges, the tree's min/max values are maintained and
> consulted
> + * in O(1) in the fastpath.
> + */
Maybe that ought not be here, doesn't seem like a fundamental design
point and would thus be better suited for a comment near where this
implementation detail is located ?
