On 10/8/2018 10:16 PM, Ed Czeck wrote:
> Fail with rte_exit() when the lack of mbufs is not recoverable.
>
> Signed-off-by: Ed Czeck <ed.cz...@atomicrules.com>
<...>
> @@ -495,6 +508,33 @@ eth_ark_rx_seed_mbufs(struct ark_rx_queue *queue)
> return 0;
> }
>
> +int
> +eth_ark_rx_seed_recovery(struct ark_rx_queue *queue,
> + uint32_t *pnb,
> + struct rte_mbuf **mbufs)
> +{
> + int status = -1;
> +
> + /* Ignore small allocation failures */
> + if (*pnb <= 64)
> + return -1;
> +
> + *pnb = 64U;
> + status = rte_pktmbuf_alloc_bulk(queue->mb_pool, mbufs, *pnb);
> + if (status != 0) {
> + PMD_DRV_LOG(ERR,
> + "ARK: Could not allocate %u mbufs from pool for RX
> queue %u;"
> + " %u free buffers remaining in queue\n",
> + *pnb, queue->queue_index,
> + queue->seed_index - queue->cons_index);
> + if (queue->seed_index - queue->cons_index < 64) {
> + rte_exit(EXIT_FAILURE,
> + "ARK: No mbufs in pool; do you have a
> leak?\n");
Hi Ed,
We are trying to get rid of rte_exit() calls, reasoning is DPDK is a library and
application should decide to exit or not, and a DPDK application may be using
multiple NICs/libraries/functionalities, failure in one of them should not cause
all application to exit.
Can you please replace rte_exit() usage?
