* Callback function to initialize mbufs for Multi-Packet RQ.
*/
static inline void
-mlx5_mprq_buf_init(struct rte_mempool *mp, void *opaque_arg __rte_unused,
+mlx5_mprq_buf_init(struct rte_mempool *mp, void *opaque_arg,
void *_m, unsigned int i __rte_unused)
{
struct mlx5_mprq_buf *buf = _m;
+ struct rte_mbuf_ext_shared_info *shinfo;
+ unsigned int strd_n = (unsigned int)(uintptr_t)opaque_arg;
+ unsigned int j;
memset(_m, 0, sizeof(*buf));
buf->mp = mp;
rte_atomic16_set(&buf->refcnt, 1);
+ for (j = 0; j != strd_n; ++j) {
+ shinfo = &buf->shinfos[j];
+ shinfo->free_cb = mlx5_mprq_buf_free_cb;
+ shinfo->fcb_opaque = buf;
+ }
}
/**
}
assert(strd_num_n && strd_sz_n);
buf_len = (1 << strd_num_n) * (1 << strd_sz_n);
- obj_size = buf_len + sizeof(struct mlx5_mprq_buf);
+ obj_size = sizeof(struct mlx5_mprq_buf) + buf_len + (1 << strd_num_n) *
+ sizeof(struct rte_mbuf_ext_shared_info) + RTE_PKTMBUF_HEADROOM;
/*
* Received packets can be either memcpy'd or externally referenced. In
* case that the packet is attached to an mbuf as an external buffer, as
}
snprintf(name, sizeof(name), "port-%u-mprq", dev->data->port_id);
mp = rte_mempool_create(name, obj_num, obj_size, MLX5_MPRQ_MP_CACHE_SZ,
- 0, NULL, NULL, mlx5_mprq_buf_init, NULL,
+ 0, NULL, NULL, mlx5_mprq_buf_init,
+ (void *)(uintptr_t)(1 << strd_num_n),
dev->device->numa_node, 0);
if (mp == NULL) {
DRV_LOG(ERR,
* Otherwise, enable Rx scatter if necessary.
*/
assert(mb_len >= RTE_PKTMBUF_HEADROOM);
- mprq_stride_size =
- dev->data->dev_conf.rxmode.max_rx_pkt_len +
- sizeof(struct rte_mbuf_ext_shared_info) +
- RTE_PKTMBUF_HEADROOM;
+ mprq_stride_size = dev->data->dev_conf.rxmode.max_rx_pkt_len +
+ RTE_PKTMBUF_HEADROOM;
if (mprq_en &&
desc > (1U << config->mprq.stride_num_n) &&
mprq_stride_size <= (1U << config->mprq.max_stride_size_n)) {
volatile struct mlx5_cqe *cqe, uint32_t rss_hash_res);
static __rte_always_inline void
-mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx);
+mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx,
+ const unsigned int strd_n);
static int
mlx5_queue_state_modify(struct rte_eth_dev *dev,
scat = &((volatile struct mlx5_wqe_mprq *)
rxq->wqes)[i].dseg;
- addr = (uintptr_t)mlx5_mprq_buf_addr(buf);
+ addr = (uintptr_t)mlx5_mprq_buf_addr(buf,
+ 1 << rxq->strd_num_n);
byte_count = (1 << rxq->strd_sz_n) *
(1 << rxq->strd_num_n);
} else {
}
static inline void
-mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx)
+mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx,
+ const unsigned int strd_n)
{
struct mlx5_mprq_buf *rep = rxq->mprq_repl;
volatile struct mlx5_wqe_data_seg *wqe =
/* Replace MPRQ buf. */
(*rxq->mprq_bufs)[rq_idx] = rep;
/* Replace WQE. */
- addr = mlx5_mprq_buf_addr(rep);
+ addr = mlx5_mprq_buf_addr(rep, strd_n);
wqe->addr = rte_cpu_to_be_64((uintptr_t)addr);
/* If there's only one MR, no need to replace LKey in WQE. */
if (unlikely(mlx5_mr_btree_len(&rxq->mr_ctrl.cache_bh) > 1))
if (consumed_strd == strd_n) {
/* Replace WQE only if the buffer is still in use. */
if (rte_atomic16_read(&buf->refcnt) > 1) {
- mprq_buf_replace(rxq, rq_ci & wq_mask);
+ mprq_buf_replace(rxq, rq_ci & wq_mask, strd_n);
/* Release the old buffer. */
mlx5_mprq_buf_free(buf);
} else if (unlikely(rxq->mprq_repl == NULL)) {
if (rxq->crc_present)
len -= RTE_ETHER_CRC_LEN;
offset = strd_idx * strd_sz + strd_shift;
- addr = RTE_PTR_ADD(mlx5_mprq_buf_addr(buf), offset);
+ addr = RTE_PTR_ADD(mlx5_mprq_buf_addr(buf, strd_n), offset);
/* Initialize the offload flag. */
pkt->ol_flags = 0;
/*
*/
buf_iova = rte_mempool_virt2iova(buf) +
RTE_PTR_DIFF(addr, buf);
- shinfo = rte_pktmbuf_ext_shinfo_init_helper(addr,
- &buf_len, mlx5_mprq_buf_free_cb, buf);
+ shinfo = &buf->shinfos[strd_idx];
+ rte_mbuf_ext_refcnt_set(shinfo, 1);
/*
* EXT_ATTACHED_MBUF will be set to pkt->ol_flags when
* attaching the stride to mbuf and more offload flags
struct rte_mempool *mp;
rte_atomic16_t refcnt; /* Atomically accessed refcnt. */
uint8_t pad[RTE_PKTMBUF_HEADROOM]; /* Headroom for the first packet. */
+ struct rte_mbuf_ext_shared_info shinfos[];
+ /*
+ * Shared information per stride.
+ * More memory will be allocated for the first stride head-room and for
+ * the strides data.
+ */
} __rte_cache_aligned;
/* Get pointer to the first stride. */
-#define mlx5_mprq_buf_addr(ptr) ((ptr) + 1)
+#define mlx5_mprq_buf_addr(ptr, strd_n) (RTE_PTR_ADD((ptr), \
+ sizeof(struct mlx5_mprq_buf) + \
+ (strd_n) * \
+ sizeof(struct rte_mbuf_ext_shared_info) + \
+ RTE_PKTMBUF_HEADROOM))
#define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
#define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9