Re: [PATCH net-next] net: bcmgenet: collect Rx discarded packet count

[Florian Fainelli <f.fainelli@gmail.com>]

On 10/03/15 12:18, Petri Gynther wrote:
> Bits 31:16 of RDMA_PROD_INDEX contain Rx discarded packet count, which
> are the Rx packets that had to be dropped by MAC hardware since there
> was no room on the Rx queue. Add code to collect this information into
> the netdev stats.
> 
> Signed-off-by: Petri Gynther <pgynther@google.com>
> ---
>  drivers/net/ethernet/broadcom/genet/bcmgenet.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
> index 275be56..7aa1834 100644
> --- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
> +++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
> @@ -1384,9 +1384,19 @@ static unsigned int bcmgenet_desc_rx(struct bcmgenet_priv *priv,
>  	int len, err;
>  	unsigned int rxpktprocessed = 0, rxpkttoprocess;
>  	unsigned int p_index;
> +	unsigned int discards;
>  	unsigned int chksum_ok = 0;
>  
>  	p_index = bcmgenet_rdma_ring_readl(priv, index, RDMA_PROD_INDEX);
> +
> +	discards = (p_index >> DMA_P_INDEX_DISCARD_CNT_SHIFT) &
> +		   DMA_P_INDEX_DISCARD_CNT_MASK;
> +	if (discards > 0) {
> +		bcmgenet_rdma_ring_writel(priv, index, 0, RDMA_PROD_INDEX);

This adds an expensive register write (~300ns on MIPS, ~200ns on ARM) in
the hot-path, and the counter saturation happens at 0xffff, which I
would prefer we deal with explicitly, even though that means missing a
bunch of discard events once we have already saturated, rather than
resetting this counter *and* the producer index for every NAPI round we
get called.

You could also deal with this counter in the ethtool gstats functions
and perform the saturation handle there, since this is a slow path already.

> +		dev->stats.rx_missed_errors += discards;
> +		dev->stats.rx_errors += discards;
> +	}
> +
>  	p_index &= DMA_P_INDEX_MASK;
>  
>  	if (likely(p_index >= ring->c_index))
> 


-- 
Florian