[xdp-hints] Re: [PATCH net-next 3/4] net: stmmac: add Rx HWTS metadata to XDP receive pkt

["Song, Yoong Siang" <yoong.siang.song@intel.com>]

On Wednesday, April 12, 2023 9:31 AM, Song Yoong Siang <yoong.siang.song@intel.com> wrote:
>On Tuesday, April 11, 2023 12:25 AM, Stanislav Fomichev <sdf@google.com>
>wrote:
>>On 04/10, Song Yoong Siang wrote:
>>> Add receive hardware timestamp metadata support via kfunc to XDP
>>> receive packets.
>>>
>>> Signed-off-by: Song Yoong Siang <yoong.siang.song@intel.com>
>>> ---
>>>  drivers/net/ethernet/stmicro/stmmac/stmmac.h  |  1 +
>>> .../net/ethernet/stmicro/stmmac/stmmac_main.c | 24
>>> +++++++++++++++++--
>>>  2 files changed, 23 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
>>> b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
>>> index ac8ccf851708..760445275da8 100644
>>> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
>>> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
>>> @@ -94,6 +94,7 @@ struct stmmac_rx_buffer {
>>>
>>>  struct stmmac_xdp_buff {
>>>  	struct xdp_buff xdp;
>>> +	ktime_t rx_hwts;
>>>  };
>>>
>>>  struct stmmac_rx_queue {
>>> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> index f7bbdf04d20c..ca183fbfde85 100644
>>> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> @@ -5307,6 +5307,8 @@ static int stmmac_rx(struct stmmac_priv *priv,
>>> int
>>limit, u32 queue)
>>>  			}
>>>  		}
>>>
>>
>>[..]
>>
>>> +		stmmac_get_rx_hwtstamp(priv, p, np, &ctx.rx_hwts);
>>
>>Do we want to pay this cost for every packet?
>>
>>The preferred alternative is to store enough state in the
>>stmmac_xdp_buff so we can get to this data from stmmac_xdp_rx_timestamp.
>>
>>I haven't read this code, but tentatively:
>>- move priv, p, np into stmmac_xdp_buff, assign them here instead of
>>  calling stmmac_get_rx_hwtstamp
>>- call stmmac_get_rx_hwtstamp from stmmac_xdp_rx_timestamp with the
>>  stored priv, p, np
>>
>>That would ensure that we won't waste the cycles pulling out the rx
>>timestamp for every packet if the higher levels / users don't care.
>>
>>Would something like this work?
>
>Hi Stanislav Fomichev,
>
>Thanks for your comments.
>
>Original stmmac_rx() function is already calling stmmac_get_rx_hwtstamp() for
>every packet. This patch move the calling of stmmac_get_rx_hwtstamp() earlier
>so that rx timestamp is available before running bpf_prog_run_xdp(). So, i think
>no additional cost introduced here. Any other thoughts?
>
>Furthermore, stmmac_get_rx_hwtstamp() will check whether hw timestamp is
>enabled in driver and  available in the descriptor before getting the hw timestamp.
>
Hi Stanislav Fomichev,

I think twice. It might add some latency for certain verdict if the hw timestamp is
enabled but the user app dint need it. I will take your suggestion and try to pull
the timestamp per need basic. Will submit v3 soon.

Thanks & Regards
Siang
>>
>>> +
>>>  		if (!skb) {
>>>  			unsigned int pre_len, sync_len;
>>>
>>> @@ -5315,7 +5317,7 @@ static int stmmac_rx(struct stmmac_priv *priv,
>>> int limit, u32 queue)
>>>
>>>  			xdp_init_buff(&ctx.xdp, buf_sz, &rx_q->xdp_rxq);
>>>  			xdp_prepare_buff(&ctx.xdp, page_address(buf->page),
>>> -					 buf->page_offset, buf1_len, false);
>>> +					 buf->page_offset, buf1_len, true);
>>>
>>>  			pre_len = ctx.xdp.data_end - ctx.xdp.data_hard_start -
>>>  				  buf->page_offset;
>>> @@ -5411,7 +5413,7 @@ static int stmmac_rx(struct stmmac_priv *priv,
>>> int limit, u32 queue)
>>>
>>>  		shhwtstamp = skb_hwtstamps(skb);
>>>  		memset(shhwtstamp, 0, sizeof(struct skb_shared_hwtstamps));
>>> -		stmmac_get_rx_hwtstamp(priv, p, np, &shhwtstamp->hwtstamp);
>
>Original stmmac_get_rx_hwtstamp() function is called here.
>
>Thanks & Regards
>Siang
>
>>> +		shhwtstamp->hwtstamp = ctx.rx_hwts;
>>>
>>>  		stmmac_rx_vlan(priv->dev, skb);
>>>  		skb->protocol = eth_type_trans(skb, priv->dev); @@ -7071,6
>>+7073,22
>>> @@ void stmmac_fpe_handshake(struct stmmac_priv *priv, bool enable)
>>>  	}
>>>  }
>>>
>>> +static int stmmac_xdp_rx_timestamp(const struct xdp_md *_ctx, u64
>>> +*timestamp) {
>>> +	const struct stmmac_xdp_buff *ctx = (void *)_ctx;
>>> +
>>> +	if (ctx->rx_hwts) {
>>> +		*timestamp = ctx->rx_hwts;
>>> +		return 0;
>>> +	}
>>> +
>>> +	return -ENODATA;
>>> +}
>>> +
>>> +const struct xdp_metadata_ops stmmac_xdp_metadata_ops = {
>>> +	.xmo_rx_timestamp		= stmmac_xdp_rx_timestamp,
>>> +};
>>> +
>>>  /**
>>>   * stmmac_dvr_probe
>>>   * @device: device pointer
>>> @@ -7178,6 +7196,8 @@ int stmmac_dvr_probe(struct device *device,
>>>
>>>  	ndev->netdev_ops = &stmmac_netdev_ops;
>>>
>>> +	ndev->xdp_metadata_ops = &stmmac_xdp_metadata_ops;
>>> +
>>>  	ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM |
>>NETIF_F_IPV6_CSUM |
>>>  			    NETIF_F_RXCSUM;
>>>  	ndev->xdp_features = NETDEV_XDP_ACT_BASIC |
>>NETDEV_XDP_ACT_REDIRECT
>>> |
>>> --
>>> 2.34.1
>>>