Re: [PATCH for-next v2 2/2] RDMA/efa: Add AH cache handling on create and destroy AH

[Leon Romanovsky <leon@kernel.org>]

On Tue, May 12, 2026 at 06:11:21AM +0000, Yonatan Nachum wrote:
> On create AH, first check if the AH cache entry already exists and if
> so, returns the already stored AH number. If the entry doesn't exist,
> the driver creates it and calls the device to create the AH. A per-entry
> mutex serializes concurrent device commands on the same AH cache entry,
> ensuring only one thread issues the device create while others wait and
> reuse the result. If the device create fails, the entry remains
> uninitialized so subsequent threads calls can create the AH.
> 
> On destroy AH, the refcount is checked and if it's the last reference,
> the driver issues the device destroy command while holding the entry
> mutex. The entry remains in the hashtable during destroy to allow
> concurrent create threads to find it and wait on the entry mutex,
> preventing create-before-destroy races on the device. After the device
> destroy completes, the entry is either recycled if new users arrived or
> removed and freed.
> 
> Reviewed-by: Firas Jahjah <firasj@amazon.com>
> Reviewed-by: Michael Margolin <mrgolin@amazon.com>
> Signed-off-by: Yonatan Nachum <ynachum@amazon.com>
> ---
>  drivers/infiniband/hw/efa/efa_ah_cache.c | 124 +++++++++++++++++++++++
>  drivers/infiniband/hw/efa/efa_ah_cache.h |   5 +
>  drivers/infiniband/hw/efa/efa_com_cmd.c  |  27 +++++
>  drivers/infiniband/hw/efa/efa_com_cmd.h  |   1 +
>  drivers/infiniband/hw/efa/efa_verbs.c    |   9 +-
>  5 files changed, 162 insertions(+), 4 deletions(-)

<...>

> +
> +/**
> + * efa_ah_cache_put - Release the final reference to an AH cache entry
> + * @ah_cache: AH cache
> + * @entry: AH cache entry
> + *
> + * Decrement the refcount. If it reaches zero, the entry is removed from the
> + * hashtable and freed. Otherwise, the entry is kept for reuse.
> + *
> + * Called after the device destroy completes or on a failed create to release
> + * the caller's reference.
> + */
> +void efa_ah_cache_put(struct efa_ah_cache *ah_cache, struct efa_ah_cache_entry *entry)
> +{
> +	mutex_lock(&ah_cache->lock);
> +	if (!refcount_dec_and_test(&entry->refcount)) {
> +		mutex_unlock(&ah_cache->lock);
> +		return;
> +	}
> +
> +	rhashtable_remove_fast(&ah_cache->hashtable, &entry->linkage, ah_cache_params);
> +	mutex_unlock(&ah_cache->lock);
> +
> +	mutex_destroy(&entry->lock);
> +	kvfree(entry);
> +}

This pattern looks very similar to kref_put_mutex().

Thanks