[dpdk-dev] [PATCH 1/2] net/ixgbe: calculate the correct number of received packets in bulk alloc function

Ananyev, Konstantin konstantin.ananyev at intel.com
Wed Feb 1 17:19:33 CET 2017


Hi,

> -----Original Message-----
> From: Jianbo Liu [mailto:jianbo.liu at linaro.org]
> Sent: Monday, December 19, 2016 6:09 AM
> To: dev at dpdk.org; Zhang, Helin <helin.zhang at intel.com>; Ananyev, Konstantin <konstantin.ananyev at intel.com>;
> jerin.jacob at caviumnetworks.com
> Cc: Jianbo Liu <jianbo.liu at linaro.org>
> Subject: [PATCH 1/2] net/ixgbe: calculate the correct number of received packets in bulk alloc function
> 
> To get better performance, Rx bulk alloc recv function will scan 8 descriptors
> in one time, but the statuses are not consistent on ARM platform because
> the memory allocated for Rx descriptors is cacheable hugepages.
> This patch is to calculate the number of received packets by scanning DD bit
> sequentially, and stops when meeting the first packet with DD bit unset.
> 
> Signed-off-by: Jianbo Liu <jianbo.liu at linaro.org>
> ---
>  drivers/net/ixgbe/ixgbe_rxtx.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/ixgbe/ixgbe_rxtx.c b/drivers/net/ixgbe/ixgbe_rxtx.c
> index b2d9f45..2866bdb 100644
> --- a/drivers/net/ixgbe/ixgbe_rxtx.c
> +++ b/drivers/net/ixgbe/ixgbe_rxtx.c
> @@ -1402,17 +1402,21 @@ ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
>  	for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
>  	     i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
>  		/* Read desc statuses backwards to avoid race condition */
> -		for (j = LOOK_AHEAD-1; j >= 0; --j)
> +		for (j = LOOK_AHEAD - 1; j >= 0; --j) {
>  			s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
> -
> -		for (j = LOOK_AHEAD - 1; j >= 0; --j)
>  			pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.
>  						       lo_dword.data);
> +		}
> +
> +		rte_smp_rmb();

If reads can be reordered, shouldn't we fill pkt_info[] after smp_rmb() here?
As another nit - with rmb() in and because you are looking the first gap in s[] now,
no need to read TXDs in backward order.
How it looks to me (as a suggestion):

for (j = 0; j != LOOK_AHEAD; j++)
	s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);

rte_smp_rmb();

for (j = 0; j < LOOK_AHEAD && (s[j] & IXGBE_RXDADV_STAT_DD) != 0; j++)
	;

for (j = 0; j < nb_dd; ++j) {
	pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.lo_dword.data);
               ....

Konstantin


> 
>  		/* Compute how many status bits were set */
>  		nb_dd = 0;
>  		for (j = 0; j < LOOK_AHEAD; ++j)
> -			nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
> +			if (s[j] & IXGBE_RXDADV_STAT_DD)
> +				++nb_dd;
> +			else
> +				break;
> 
>  		nb_rx += nb_dd;
> 
> --
> 2.4.11



More information about the dev mailing list