Re: [Qemu-devel] [PATCH v5 13/18] qht: support parallel writes

[Sergey Fedorov]

On 14/05/16 06:34, Emilio G. Cota wrote:
> +/*
> + * Get a head bucket and lock it, making sure its parent map is not stale.
> + * @pmap is filled with a pointer to the bucket's parent map.
> + *
> + * Unlock with qemu_spin_unlock(&b->lock).
> + */
> +static inline
> +struct qht_bucket *qht_bucket_lock__no_stale(struct qht *ht, uint32_t hash,
> +                                             struct qht_map **pmap)
> +{
> +    struct qht_bucket *b;
> +    struct qht_map *map;
> +
> +    for (;;) {
> +        map = qht_map__atomic_mb(ht);
> +        b = qht_map_to_bucket(map, hash);
> +
> +        qemu_spin_lock(&b->lock);
> +        if (likely(!map->stale)) {
> +            *pmap = map;
> +            return b;
> +        }
> +        qemu_spin_unlock(&b->lock);
> +
> +        /* resize in progress; wait until it completes */
> +        while (qemu_spin_locked(&ht->lock)) {
> +            cpu_relax();
> +        }
> +    }
> +}

What if we turn qht::lock into a mutex and change the function as follows:

    static inline
    struct qht_bucket *qht_bucket_lock__no_stale(struct qht *ht,
    uint32_t hash,
                                                 struct qht_map
    **pmap)       
    {
        struct qht_bucket *b;
        struct qht_map *map;

        map = atomic_rcu_read(&ht->map);
        b = qht_map_to_bucket(map, hash);

        qemu_spin_lock(&b->lock);
        /* 'ht->map' access is serialized by 'b->lock' here */
        if (likely(map == ht->map)) {
            /* no resize in progress; we're done */
            *pmap = map;
            return b;
        }
        qemu_spin_unlock(&b->lock);

        /* resize in progress; retry grabbing 'ht->lock' */
        qemu_mutex_lock(&ht->lock);
        b = qht_map_to_bucket(ht->map, hash);
        *pmap = ht->map;
        qemu_spin_lock(&b->lock);
        qemu_mutex_unlock(&ht->lock);

        return b;
    }


With this implementation we could:
 (1) get rid of qht_map::stale
 (2) don't waste cycles waiting for resize to complete

Kind regards,
Sergey