Re: [PATCH rdma-next] RDMA/irdma: Add support for address handle re-use

[Jason Gunthorpe <jgg@nvidia.com>]

On Thu, Jan 20, 2022 at 11:40:41AM -0600, Shiraz Saleem wrote:
> +/**
> + * irdma_ah_exists - Check for existing identical AH
> + * @iwdev: irdma device
> + * @new_ah: AH to check for
> + *
> + * returns true if AH is found, false if not found.
> + */
> +static bool irdma_ah_exists(struct irdma_device *iwdev,
> +			    struct irdma_ah *new_ah)
> +{
> +	struct irdma_ah *ah;
>  
> -		if (!cnt) {
> -			ibdev_dbg(&iwdev->ibdev,
> -				  "VERBS: CQP create AH timed out");
> -			err = -ETIMEDOUT;
> -			goto error;
> +	list_for_each_entry (ah, &iwdev->ah_list, list) {
> +		/* Set ah_valid and ah_id the same so memcmp can work */
> +		new_ah->sc_ah.ah_info.ah_idx = ah->sc_ah.ah_info.ah_idx;
> +		new_ah->sc_ah.ah_info.ah_valid = ah->sc_ah.ah_info.ah_valid;
> +		if (!memcmp(&ah->sc_ah.ah_info, &new_ah->sc_ah.ah_info,
> +			    sizeof(ah->sc_ah.ah_info))) {
> +			refcount_inc(&ah->refcnt);
> +			new_ah->parent_ah = ah;
> +			return true;
>  		}
>  	}

So, the number of AHs is so large the HW has problems but you propose
to use a linear search to de-dup them?

Are you sure?

> +static int irdma_create_user_ah(struct ib_ah *ibah,
> +				struct rdma_ah_init_attr *attr,
> +				struct ib_udata *udata)
> +{
> +	struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
> +	struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
> +	struct irdma_create_ah_resp uresp;
> +	struct irdma_ah *parent_ah;
> +	int err;
> +
> +	err = irdma_setup_ah(ibah, attr);
> +	if (err)
> +		return err;
> +	if (attr->flags & RDMA_CREATE_AH_SLEEPABLE) {
> +		mutex_lock(&iwdev->ah_list_lock);

User AH's are always sleepable, no need for these extra paths.

Jason