Re: [Intel-wired-lan] [PATCH iwl-next v4] idpf: export RX hardware timestamping information to XDP

[Paul Menzel <pmenzel@molgen.mpg.de>]

Dear Mina,


Thank you for your patch. Some minor comments, should you resend.

Am 23.12.25 um 20:46 schrieb Mina Almasry via Intel-wired-lan:
> From: YiFei Zhu <zhuyifei@google.com>
> 
> The logic is similar to idpf_rx_hwtstamp, but the data is exported
> as a BPF kfunc instead of appended to an skb to support grabbing
> timestamps in xsk packets.
> 
> A idpf_queue_has(PTP, rxq) condition is added to check the queue
> supports PTP similar to idpf_rx_process_skb_fields.
> 
> Tested using an xsk connection and checking xdp timestamps are
> retreivable in received packets.

retr*ie*vable

It’d be great if you could share the commands.

> Cc: intel-wired-lan@lists.osuosl.org
> Signed-off-by: YiFei Zhu <zhuyifei@google.com>
> Signed-off-by: Mina Almasry <almasrymina@google.com>
> Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
> 
> ---
> 
> v4: https://lore.kernel.org/netdev/20251219202957.2309698-1-almasrymina@google.com/
> - Fix indentation (lobakin)
> - I kept the (u64) casts for all bit shifted bits in idpf_xdp_get_qw3
>    and friends as I see all idpf_xdp_get_qw* functions do the cast in all
>    bit-shifted variables.
> 
> v3: https://lore.kernel.org/netdev/20251218022948.3288897-1-almasrymina@google.com/
> - Do the idpf_queue_has(PTP) check before we read qw1 (lobakin)
> - Fix _qw1 not copying over ts_low on on !__LIBETH_WORD_ACCESS systems
>    (AI)
> 
> v2: https://lore.kernel.org/netdev/20251122140839.3922015-1-almasrymina@google.com/
> - Fixed alphabetical ordering
> - Use the xdp desc type instead of virtchnl one (required some added
>    helpers)
> 
> ---
>   drivers/net/ethernet/intel/idpf/xdp.c | 31 +++++++++++++++++++++++++++
>   drivers/net/ethernet/intel/idpf/xdp.h | 20 +++++++++++++++++
>   2 files changed, 51 insertions(+)
> 
> diff --git a/drivers/net/ethernet/intel/idpf/xdp.c b/drivers/net/ethernet/intel/idpf/xdp.c
> index 958d16f87424..0916d201bf98 100644
> --- a/drivers/net/ethernet/intel/idpf/xdp.c
> +++ b/drivers/net/ethernet/intel/idpf/xdp.c
> @@ -2,6 +2,7 @@
>   /* Copyright (C) 2025 Intel Corporation */
>   
>   #include "idpf.h"
> +#include "idpf_ptp.h"
>   #include "idpf_virtchnl.h"
>   #include "xdp.h"
>   #include "xsk.h"
> @@ -391,8 +392,38 @@ static int idpf_xdpmo_rx_hash(const struct xdp_md *ctx, u32 *hash,
>   				    pt);
>   }
>   
> +static int idpf_xdpmo_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
> +{
> +	const struct libeth_xdp_buff *xdp = (typeof(xdp))ctx;
> +	struct idpf_xdp_rx_desc desc __uninitialized;
> +	const struct idpf_rx_queue *rxq;
> +	u64 cached_time, ts_ns;
> +	u32 ts_high;
> +
> +	rxq = libeth_xdp_buff_to_rq(xdp, typeof(*rxq), xdp_rxq);
> +
> +	if (!idpf_queue_has(PTP, rxq))
> +		return -ENODATA;
> +
> +	idpf_xdp_get_qw1(&desc, xdp->desc);
> +
> +	if (!(idpf_xdp_rx_ts_low(&desc) & VIRTCHNL2_RX_FLEX_TSTAMP_VALID))
> +		return -ENODATA;
> +
> +	cached_time = READ_ONCE(rxq->cached_phc_time);
> +
> +	idpf_xdp_get_qw3(&desc, xdp->desc);
> +
> +	ts_high = idpf_xdp_rx_ts_high(&desc);
> +	ts_ns = idpf_ptp_tstamp_extend_32b_to_64b(cached_time, ts_high);
> +
> +	*timestamp = ts_ns;
> +	return 0;
> +}
> +
>   static const struct xdp_metadata_ops idpf_xdpmo = {
>   	.xmo_rx_hash		= idpf_xdpmo_rx_hash,
> +	.xmo_rx_timestamp	= idpf_xdpmo_rx_timestamp,

Append the unit?

>   };
>   
>   void idpf_xdp_set_features(const struct idpf_vport *vport)
> diff --git a/drivers/net/ethernet/intel/idpf/xdp.h b/drivers/net/ethernet/intel/idpf/xdp.h
> index 479f5ef3c604..1748a0d73547 100644
> --- a/drivers/net/ethernet/intel/idpf/xdp.h
> +++ b/drivers/net/ethernet/intel/idpf/xdp.h
> @@ -112,11 +112,13 @@ struct idpf_xdp_rx_desc {
>   	aligned_u64		qw1;
>   #define IDPF_XDP_RX_BUF		GENMASK_ULL(47, 32)
>   #define IDPF_XDP_RX_EOP		BIT_ULL(1)
> +#define IDPF_XDP_RX_TS_LOW	GENMASK_ULL(31, 24)
>   
>   	aligned_u64		qw2;
>   #define IDPF_XDP_RX_HASH	GENMASK_ULL(31, 0)
>   
>   	aligned_u64		qw3;
> +#define IDPF_XDP_RX_TS_HIGH	GENMASK_ULL(63, 32)
>   } __aligned(4 * sizeof(u64));
>   static_assert(sizeof(struct idpf_xdp_rx_desc) ==
>   	      sizeof(struct virtchnl2_rx_flex_desc_adv_nic_3));
> @@ -128,6 +130,8 @@ static_assert(sizeof(struct idpf_xdp_rx_desc) ==
>   #define idpf_xdp_rx_buf(desc)	FIELD_GET(IDPF_XDP_RX_BUF, (desc)->qw1)
>   #define idpf_xdp_rx_eop(desc)	!!((desc)->qw1 & IDPF_XDP_RX_EOP)
>   #define idpf_xdp_rx_hash(desc)	FIELD_GET(IDPF_XDP_RX_HASH, (desc)->qw2)
> +#define idpf_xdp_rx_ts_low(desc)	FIELD_GET(IDPF_XDP_RX_TS_LOW, (desc)->qw1)
> +#define idpf_xdp_rx_ts_high(desc)	FIELD_GET(IDPF_XDP_RX_TS_HIGH, (desc)->qw3)
>   
>   static inline void
>   idpf_xdp_get_qw0(struct idpf_xdp_rx_desc *desc,
> @@ -149,6 +153,9 @@ idpf_xdp_get_qw1(struct idpf_xdp_rx_desc *desc,
>   	desc->qw1 = ((const typeof(desc))rxd)->qw1;
>   #else
>   	desc->qw1 = ((u64)le16_to_cpu(rxd->buf_id) << 32) |
> +		    ((u64)rxd->ts_low << 24) |
> +		    ((u64)rxd->fflags1 << 16) |
> +		    ((u64)rxd->status_err1 << 8) |
>   		    rxd->status_err0_qw1;
>   #endif
>   }
> @@ -166,6 +173,19 @@ idpf_xdp_get_qw2(struct idpf_xdp_rx_desc *desc,
>   #endif
>   }
>   
> +static inline void
> +idpf_xdp_get_qw3(struct idpf_xdp_rx_desc *desc,
> +		 const struct virtchnl2_rx_flex_desc_adv_nic_3 *rxd)
> +{
> +#ifdef __LIBETH_WORD_ACCESS
> +	desc->qw3 = ((const typeof(desc))rxd)->qw3;
> +#else
> +	desc->qw3 = ((u64)le32_to_cpu(rxd->ts_high) << 32) |
> +		    ((u64)le16_to_cpu(rxd->fmd6) << 16) |
> +		    le16_to_cpu(rxd->l2tag1);
> +#endif

It’s done elsewhere in the file, but I wonder why use the preprocessor 
and not plain C code, and let the linker(?) remove the unneeded branch?

> +}
> +
>   void idpf_xdp_set_features(const struct idpf_vport *vport);
>   
>   int idpf_xdp(struct net_device *dev, struct netdev_bpf *xdp);


Kind regards,

Paul