[xdp-hints] Re: [RFC bpf-next 07/23] ice: Support RX hash XDP hint

[Alexander Lobakin <aleksander.lobakin@intel.com>]

From: Larysa Zaremba <larysa.zaremba@intel.com>
Date: Thu, 24 Aug 2023 21:26:46 +0200

> RX hash XDP hint requests both hash value and type.
> Type is XDP-specific, so we need a separate way to map
> these values to the hardware ptypes, so create a lookup table.
> 
> Instead of creating a new long list, reuse contents
> of ice_decode_rx_desc_ptype[] through preprocessor.
> 
> Current hash type enum does not contain ICMP packet type,
> but ice devices support it, so also add a new type into core code.
> 
> Then use previously refactored code and create a function
> that allows XDP code to read RX hash.
> 
> Signed-off-by: Larysa Zaremba <larysa.zaremba@intel.com>

[...]

>  	/* unused entries */
> -	[154 ... 1023] = { 0, 0, 0, 0, 0, 0, 0, 0, 0 }
> +	[ICE_NUM_DEFINED_PTYPES ... 1023] = { 0, 0, 0, 0, 0, 0, 0, 0, 0 }
>  };
>  
>  static inline struct ice_rx_ptype_decoded ice_decode_rx_desc_ptype(u16 ptype)
> diff --git a/drivers/net/ethernet/intel/ice/ice_txrx_lib.c b/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> index 463d9e5cbe05..b11cfaedb81c 100644
> --- a/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> +++ b/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> @@ -567,6 +567,79 @@ static int ice_xdp_rx_hw_ts(const struct xdp_md *ctx, u64 *ts_ns)
>  	return 0;
>  }
>  
> +/* Define a ptype index -> XDP hash type lookup table.
> + * It uses the same ptype definitions as ice_decode_rx_desc_ptype[],
> + * avoiding possible copy-paste errors.
> + */
> +#undef ICE_PTT
> +#undef ICE_PTT_UNUSED_ENTRY
> +
> +#define ICE_PTT(PTYPE, OUTER_IP, OUTER_IP_VER, OUTER_FRAG, T, TE, TEF, I, PL)\
> +	[PTYPE] = XDP_RSS_L3_##OUTER_IP_VER | XDP_RSS_L4_##I | XDP_RSS_TYPE_##PL
> +
> +#define ICE_PTT_UNUSED_ENTRY(PTYPE) [PTYPE] = 0
> +
> +/* A few supplementary definitions for when XDP hash types do not coincide
> + * with what can be generated from ptype definitions
> + * by means of preprocessor concatenation.
> + */
> +#define XDP_RSS_L3_NONE		XDP_RSS_TYPE_NONE
> +#define XDP_RSS_L4_NONE		XDP_RSS_TYPE_NONE
> +#define XDP_RSS_TYPE_PAY2	XDP_RSS_TYPE_L2
> +#define XDP_RSS_TYPE_PAY3	XDP_RSS_TYPE_NONE
> +#define XDP_RSS_TYPE_PAY4	XDP_RSS_L4
> +
> +static const enum xdp_rss_hash_type
> +ice_ptype_to_xdp_hash[ICE_NUM_DEFINED_PTYPES] = {
> +	ICE_PTYPES
> +};

Is there a big win in performance with this 600-byte static table
comparing to having several instructions which would do
to_parsed_ptype() and then build a return enum according to its fields?
I believe that would cost only several instructions. Not that it's a
disaster to consume 600 more bytes of rodata, but still.

Alternatively, you can look at how parsed ptype is compressed to 16 bit
in libie and use those saved bits to encode complete XDP RSS hash enum
directly there, so that ice_ptype_lkup[] would have both parsed ptype
and XDP hash return value :D

> +
> +#undef XDP_RSS_L3_NONE
> +#undef XDP_RSS_L4_NONE
> +#undef XDP_RSS_TYPE_PAY2
> +#undef XDP_RSS_TYPE_PAY3
> +#undef XDP_RSS_TYPE_PAY4

[...]

Thanks,
Olek