RE: [PATCH v2] NTB: allocate number transport entries depending on size of ring size

["Allen Hubbe" <Allen.Hubbe@emc.com>]

> From: Dave Jiang
> Currently we only allocate a fixed default number of descriptors for the tx
> and rx side. We should dynamically resize it to the number of descriptors
> resides in the transport rings. We should know the number of transmit
> descriptors at initializaiton. We will allocate the default number of
> descriptors for receive side and allocate additional ones when we know the
> actual max entries for receive.
> 
> Signed-off-by: Dave Jiang <dave.jiang@intel.com>
> ---
>  drivers/ntb/ntb_transport.c |   25 ++++++++++++++++++++++++-
>  1 file changed, 24 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
> index 2ef9d913..a943bb1 100644
> --- a/drivers/ntb/ntb_transport.c
> +++ b/drivers/ntb/ntb_transport.c
> @@ -597,9 +597,13 @@ static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt,
>  {
>  	struct ntb_transport_qp *qp = &nt->qp_vec[qp_num];
>  	struct ntb_transport_mw *mw;
> +	struct ntb_dev *ndev = nt->ndev;
> +	struct ntb_queue_entry *entry;
>  	unsigned int rx_size, num_qps_mw;
>  	unsigned int mw_num, mw_count, qp_count;
>  	unsigned int i;
> +	unsigned int rx_entries = 0;
> +	int node;
> 
>  	mw_count = nt->mw_count;
>  	qp_count = nt->qp_count;
> @@ -626,6 +630,25 @@ static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt,
>  	qp->rx_max_entry = rx_size / qp->rx_max_frame;
>  	qp->rx_index = 0;
> 
> +	/*
> +	 * Checking to see if we have more entries than the default.
> +	 * We should add additional entries if that is the case so we
> +	 * can be in sync with the transport frames.
> +	 */
> +	if (qp->rx_max_entry > NTB_QP_DEF_NUM_ENTRIES)
> +		rx_entries = qp->rx_max_entry - NTB_QP_DEF_NUM_ENTRIES;

If the link drops and then returns, it looks like this will we allocate
another (rx_max_entries - DEFAULT) each time we establish the link.
Do we have an actual count of the number of entries that we could use
here, instead of NTB_QP_DEF_NUM_ENTRIES?

If we don't already track the actual number of entries allocated,
different from the maximum number of entries that fit in the ring
buffer, maybe we should add it.  Something like qp->rx_num_entry?

It is possible, if the memory window is small or mtu is large, we have
allocated more entries than max.  The term max is deceptive, oh well.

> +
> +	node = dev_to_node(&ndev->dev);
> +	for (i = 0; i < rx_entries; i++) {

Instead of:
if (num_entry < max_entry)
	calc the difference;
for (i = 0 .. the difference) {
        /* do alloc */
}

What about:

for (i = num_entry; i < max_entry; ++i) {
        /* do alloc */
}
current_entry = max_entry;

or

while (current_entry < max_entry) {
        /* do alloc */
        ++current_entry;
}

> +		entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
> +		if (!entry)
> +			return -ENOMEM;
> +
> +		entry->qp = qp;
> +		ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry,
> +			     &qp->rx_free_q);
> +	}
> +
>  	qp->remote_rx_info->entry = qp->rx_max_entry - 1;
> 
>  	/* setup the hdr offsets with 0's */
> @@ -1723,7 +1746,7 @@ ntb_transport_create_queue(void *data, struct device *client_dev,
>  			     &qp->rx_free_q);
>  	}
> 
> -	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
> +	for (i = 0; i < qp->tx_max_entry; i++) {
>  		entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
>  		if (!entry)
>  			goto err2;