Re: [PATCH net-next 5/8] net: lantiq: configure the burst length in ethernet drivers

From: Aleksander Bajkowski
Date: Sun Sep 19 2021 - 14:16:23 EST


Hi Hauke,

On 9/15/21 12:36 AM, Hauke Mehrtens wrote:
> On 9/14/21 11:21 PM, Aleksander Jan Bajkowski wrote:
>> Configure the burst length in Ethernet drivers. This improves
>> Ethernet performance by 58%. According to the vendor BSP,
>> 8W burst length is supported by ar9 and newer SoCs.
>>
>> The NAT benchmark results on xRX200 (Down/Up):
>> * 2W: 330 Mb/s
>> * 4W: 432 Mb/s    372 Mb/s
>> * 8W: 520 Mb/s    389 Mb/s
>>
>> Tested on xRX200 and xRX330.
>>
>> Signed-off-by: Aleksander Jan Bajkowski <olek2@xxxxx>
>> ---
>>   drivers/net/ethernet/lantiq_etop.c   | 21 ++++++++++++++++++---
>>   drivers/net/ethernet/lantiq_xrx200.c | 21 ++++++++++++++++++---
>>   2 files changed, 36 insertions(+), 6 deletions(-)
>>
> .....
>> diff --git a/drivers/net/ethernet/lantiq_xrx200.c b/drivers/net/ethernet/lantiq_xrx200.c
>> index fb78f17d734f..5d96248ce83b 100644
>> --- a/drivers/net/ethernet/lantiq_xrx200.c
>> +++ b/drivers/net/ethernet/lantiq_xrx200.c
>> @@ -71,6 +71,9 @@ struct xrx200_priv {
>>       struct net_device *net_dev;
>>       struct device *dev;
>>   +    int tx_burst_len;
>> +    int rx_burst_len;
>> +
>>       __iomem void *pmac_reg;
>>   };
>>   @@ -316,8 +319,8 @@ static netdev_tx_t xrx200_start_xmit(struct sk_buff *skb,
>>       if (unlikely(dma_mapping_error(priv->dev, mapping)))
>>           goto err_drop;
>>   -    /* dma needs to start on a 16 byte aligned address */
>> -    byte_offset = mapping % 16;
>> +    /* dma needs to start on a burst length value aligned address */
>> +    byte_offset = mapping % (priv->tx_burst_len * 4);
>>         desc->addr = mapping - byte_offset;
>>       /* Make sure the address is written before we give it to HW */
>> @@ -369,7 +372,7 @@ static int xrx200_dma_init(struct xrx200_priv *priv)
>>       int ret = 0;
>>       int i;
>>   -    ltq_dma_init_port(DMA_PORT_ETOP);
>> +    ltq_dma_init_port(DMA_PORT_ETOP, priv->tx_burst_len, rx_burst_len);
>>         ch_rx->dma.nr = XRX200_DMA_RX;
>>       ch_rx->dma.dev = priv->dev;
>> @@ -478,6 +481,18 @@ static int xrx200_probe(struct platform_device *pdev)
>>       if (err)
>>           eth_hw_addr_random(net_dev);
>>   +    err = device_property_read_u32(dev, "lantiq,tx-burst-length", &priv->tx_burst_len);
>> +    if (err < 0) {
>> +        dev_err(dev, "unable to read tx-burst-length property\n");
>> +        return err;
>> +    }
>> +
>> +    err = device_property_read_u32(dev, "lantiq,rx-burst-length", &priv->rx_burst_len);
>> +    if (err < 0) {
>> +        dev_err(dev, "unable to read rx-burst-length property\n");
>> +        return err;
>> +    }
>> +
>
> I would prefer if you would hard code these values to 8 for the xrx200 driver. All SoCs with this IP block should support this.
OK. I can hard code 8W burst length in the driver for xrx200. Burst length as a configurable parameter is really only needed in the lantiq_etop driver.
>
>>       /* bring up the dma engine and IP core */
>>       err = xrx200_dma_init(priv);
>>       if (err)
>>
>
> Hauke
Aleksander