[dpdk-dev] [PATCH v4 10/15] fm10k: add receive and tranmit function

Chen, Jing D jing.d.chen at intel.com
Thu Feb 12 05:04:06 CET 2015



> -----Original Message-----
> From: Shaw, Jeffrey B
> Sent: Thursday, February 12, 2015 1:29 AM
> To: Chen, Jing D
> Cc: dev at dpdk.org; Zhang, Helin; Qiu, Michael; nhorman at tuxdriver.com;
> thomas.monjalon at 6wind.com; david.marchand at 6wind.com
> Subject: Re: [PATCH v4 10/15] fm10k: add receive and tranmit function
> 
> On Wed, Feb 11, 2015 at 09:31:33AM +0800, Chen Jing D(Mark) wrote:
> 
> > +uint16_t
> > +fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> > +	uint16_t nb_pkts)
> > +{
> > +	struct rte_mbuf *mbuf;
> > +	union fm10k_rx_desc desc;
> > +	struct fm10k_rx_queue *q = rx_queue;
> > +	uint16_t count = 0;
> > +	int alloc = 0;
> > +	uint16_t next_dd;
> > +
> > +	next_dd = q->next_dd;
> > +
> > +	nb_pkts = RTE_MIN(nb_pkts, q->alloc_thresh);
> > +	for (count = 0; count < nb_pkts; ++count) {
> > +		mbuf = q->sw_ring[next_dd];
> > +		desc = q->hw_ring[next_dd];
> > +		if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
> > +			break;
> > +#ifdef RTE_LIBRTE_FM10K_DEBUG_RX
> > +		dump_rxd(&desc);
> > +#endif
> > +		rte_pktmbuf_pkt_len(mbuf) = desc.w.length;
> > +		rte_pktmbuf_data_len(mbuf) = desc.w.length;
> > +
> > +		mbuf->ol_flags = 0;
> > +#ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
> > +		rx_desc_to_ol_flags(mbuf, &desc);
> > +#endif
> > +
> > +		mbuf->hash.rss = desc.d.rss;
> > +
> > +		rx_pkts[count] = mbuf;
> > +		if (++next_dd == q->nb_desc) {
> > +			next_dd = 0;
> > +			alloc = 1;
> > +		}
> > +
> > +		/* Prefetch next mbuf while processing current one. */
> > +		rte_prefetch0(q->sw_ring[next_dd]);
> > +
> > +		/*
> > +		 * When next RX descriptor is on a cache-line boundary,
> > +		 * prefetch the next 4 RX descriptors and the next 8 pointers
> > +		 * to mbufs.
> > +		 */
> > +		if ((next_dd & 0x3) == 0) {
> > +			rte_prefetch0(&q->hw_ring[next_dd]);
> > +			rte_prefetch0(&q->sw_ring[next_dd]);
> > +		}
> > +	}
> > +
> > +	q->next_dd = next_dd;
> > +
> > +	if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
> > +		rte_mempool_get_bulk(q->mp, (void **)&q->sw_ring[q-
> >next_alloc],
> > +					q->alloc_thresh);
> 
> The return value should be checked here in case the mempool runs out
> of buffers. Thanks Helin for spotting this.  I'm not sure how I missed it
> originally.

Thanks! I'll change accordingly.

> 
> > +		for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
> > +			mbuf = q->sw_ring[q->next_alloc];
> > +
> > +			/* setup static mbuf fields */
> > +			fm10k_pktmbuf_reset(mbuf, q->port_id);
> > +
> > +			/* write descriptor */
> > +			desc.q.pkt_addr =
> MBUF_DMA_ADDR_DEFAULT(mbuf);
> > +			desc.q.hdr_addr =
> MBUF_DMA_ADDR_DEFAULT(mbuf);
> > +			q->hw_ring[q->next_alloc] = desc;
> > +		}
> > +		FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
> > +		q->next_trigger += q->alloc_thresh;
> > +		if (q->next_trigger >= q->nb_desc) {
> > +			q->next_trigger = q->alloc_thresh - 1;
> > +			q->next_alloc = 0;
> > +		}
> > +	}
> > +
> > +	return count;
> > +}
> > +
> 
> Thanks,
> Jeff


More information about the dev mailing list