[xdp-hints] Re: [PATCH bpf-next V2 1/5] igc: enable and fix RX hash usage by netstack

[Jesper Dangaard Brouer <jbrouer@redhat.com>]

On 23/04/2023 16.46, John Fastabend wrote:
> Jesper Dangaard Brouer wrote:
>> When function igc_rx_hash() was introduced in v4.20 via commit 0507ef8a0372
>> ("igc: Add transmit and receive fastpath and interrupt handlers"), the
>> hardware wasn't configured to provide RSS hash, thus it made sense to not
>> enable net_device NETIF_F_RXHASH feature bit.
>>
>> The NIC hardware was configured to enable RSS hash info in v5.2 via commit
>> 2121c2712f82 ("igc: Add multiple receive queues control supporting"), but
>> forgot to set the NETIF_F_RXHASH feature bit.
>>
>> The original implementation of igc_rx_hash() didn't extract the associated
>> pkt_hash_type, but statically set PKT_HASH_TYPE_L3. The largest portions of
>> this patch are about extracting the RSS Type from the hardware and mapping
>> this to enum pkt_hash_types. This was based on Foxville i225 software user
>> manual rev-1.3.1 and tested on Intel Ethernet Controller I225-LM (rev 03).
>>
>> For UDP it's worth noting that RSS (type) hashing have been disabled both for
>> IPv4 and IPv6 (see IGC_MRQC_RSS_FIELD_IPV4_UDP + IGC_MRQC_RSS_FIELD_IPV6_UDP)
>> because hardware RSS doesn't handle fragmented pkts well when enabled (can
>> cause out-of-order). This results in PKT_HASH_TYPE_L3 for UDP packets, and
>> hash value doesn't include UDP port numbers. Not being PKT_HASH_TYPE_L4, have
>> the effect that netstack will do a software based hash calc calling into
>> flow_dissect, but only when code calls skb_get_hash(), which doesn't
>> necessary happen for local delivery.
>>
>> For QA verification testing I wrote a small bpftrace prog:
>>   [0] https://github.com/xdp-project/xdp-project/blob/master/areas/hints/monitor_skb_hash_on_dev.bt
>>
>> Fixes: 2121c2712f82 ("igc: Add multiple receive queues control supporting")
>> Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
>> ---
>>   drivers/net/ethernet/intel/igc/igc.h      |   28 ++++++++++++++++++++++++++
>>   drivers/net/ethernet/intel/igc/igc_main.c |   31 +++++++++++++++++++++++++----
>>   2 files changed, 55 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/intel/igc/igc.h b/drivers/net/ethernet/intel/igc/igc.h
>> index 34aebf00a512..f7f9e217e7b4 100644
>> --- a/drivers/net/ethernet/intel/igc/igc.h
>> +++ b/drivers/net/ethernet/intel/igc/igc.h
>> @@ -13,6 +13,7 @@
>>   #include <linux/ptp_clock_kernel.h>
>>   #include <linux/timecounter.h>
>>   #include <linux/net_tstamp.h>
>> +#include <linux/bitfield.h>
>>   
>>   #include "igc_hw.h"
>>   
>> @@ -311,6 +312,33 @@ extern char igc_driver_name[];
>>   #define IGC_MRQC_RSS_FIELD_IPV4_UDP	0x00400000
>>   #define IGC_MRQC_RSS_FIELD_IPV6_UDP	0x00800000
>>   
>> +/* RX-desc Write-Back format RSS Type's */
>> +enum igc_rss_type_num {
>> +	IGC_RSS_TYPE_NO_HASH		= 0,
>> +	IGC_RSS_TYPE_HASH_TCP_IPV4	= 1,
>> +	IGC_RSS_TYPE_HASH_IPV4		= 2,
>> +	IGC_RSS_TYPE_HASH_TCP_IPV6	= 3,
>> +	IGC_RSS_TYPE_HASH_IPV6_EX	= 4,
>> +	IGC_RSS_TYPE_HASH_IPV6		= 5,
>> +	IGC_RSS_TYPE_HASH_TCP_IPV6_EX	= 6,
>> +	IGC_RSS_TYPE_HASH_UDP_IPV4	= 7,
>> +	IGC_RSS_TYPE_HASH_UDP_IPV6	= 8,
>> +	IGC_RSS_TYPE_HASH_UDP_IPV6_EX	= 9,
>> +	IGC_RSS_TYPE_MAX		= 10,
>> +};
>> +#define IGC_RSS_TYPE_MAX_TABLE		16
>> +#define IGC_RSS_TYPE_MASK		GENMASK(3,0) /* 4-bits (3:0) = mask 0x0F */
>> +
>> +/* igc_rss_type - Rx descriptor RSS type field */
>> +static inline u32 igc_rss_type(const union igc_adv_rx_desc *rx_desc)
>> +{
>> +	/* RSS Type 4-bits (3:0) number: 0-9 (above 9 is reserved)
>> +	 * Accessing the same bits via u16 (wb.lower.lo_dword.hs_rss.pkt_info)
>> +	 * is slightly slower than via u32 (wb.lower.lo_dword.data)
>> +	 */
>> +	return le32_get_bits(rx_desc->wb.lower.lo_dword.data, IGC_RSS_TYPE_MASK);
>> +}
>> +
>>   /* Interrupt defines */
>>   #define IGC_START_ITR			648 /* ~6000 ints/sec */
>>   #define IGC_4K_ITR			980
>> diff --git a/drivers/net/ethernet/intel/igc/igc_main.c b/drivers/net/ethernet/intel/igc/igc_main.c
>> index 1c4676882082..bfa9768d447f 100644
>> --- a/drivers/net/ethernet/intel/igc/igc_main.c
>> +++ b/drivers/net/ethernet/intel/igc/igc_main.c
>> @@ -1690,14 +1690,36 @@ static void igc_rx_checksum(struct igc_ring *ring,
>>   		   le32_to_cpu(rx_desc->wb.upper.status_error));
>>   }
>>   
>> +/* Mapping HW RSS Type to enum pkt_hash_types */
>> +static const enum pkt_hash_types igc_rss_type_table[IGC_RSS_TYPE_MAX_TABLE] = {
>> +	[IGC_RSS_TYPE_NO_HASH]		= PKT_HASH_TYPE_L2,
>> +	[IGC_RSS_TYPE_HASH_TCP_IPV4]	= PKT_HASH_TYPE_L4,
>> +	[IGC_RSS_TYPE_HASH_IPV4]	= PKT_HASH_TYPE_L3,
>> +	[IGC_RSS_TYPE_HASH_TCP_IPV6]	= PKT_HASH_TYPE_L4,
>> +	[IGC_RSS_TYPE_HASH_IPV6_EX]	= PKT_HASH_TYPE_L3,
>> +	[IGC_RSS_TYPE_HASH_IPV6]	= PKT_HASH_TYPE_L3,
>> +	[IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = PKT_HASH_TYPE_L4,
>> +	[IGC_RSS_TYPE_HASH_UDP_IPV4]	= PKT_HASH_TYPE_L4,
>> +	[IGC_RSS_TYPE_HASH_UDP_IPV6]	= PKT_HASH_TYPE_L4,
>> +	[IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = PKT_HASH_TYPE_L4,
>> +	[10] = PKT_HASH_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW  */
>> +	[11] = PKT_HASH_TYPE_NONE, /* keep array sized for SW bit-mask   */
>> +	[12] = PKT_HASH_TYPE_NONE, /* to handle future HW revisons       */
>> +	[13] = PKT_HASH_TYPE_NONE,
>> +	[14] = PKT_HASH_TYPE_NONE,
>> +	[15] = PKT_HASH_TYPE_NONE,
>> +};
>> +
>>   static inline void igc_rx_hash(struct igc_ring *ring,
>>   			       union igc_adv_rx_desc *rx_desc,
>>   			       struct sk_buff *skb)
>>   {
>> -	if (ring->netdev->features & NETIF_F_RXHASH)
>> -		skb_set_hash(skb,
>> -			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
>> -			     PKT_HASH_TYPE_L3);
>> +	if (ring->netdev->features & NETIF_F_RXHASH) {
>> +		u32 rss_hash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
>> +		u32 rss_type = igc_rss_type(rx_desc);
>> +
>> +		skb_set_hash(skb, rss_hash, igc_rss_type_table[rss_type]);
> 
> Just curious why not copy the logic from the other driver fms10k, ice, ect.
> 
> 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
> 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
> 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);

Detail: This code mis-categorize (e.g. ARP) PKT_HASH_TYPE_L2 as
PKT_HASH_TYPE_L3, but as core reduces this further to one SKB bit, it
doesn't really matter.

> avoiding the table logic. Do the driver folks care?

The define IXGBE_RSS_L4_TYPES_MASK becomes the "table" logic as a 1-bit
true/false table.  It is a more compact table, let me know if this is
preferred.

Yes, it is really upto driver maintainer people to decide, what code is
preferred ?

--Jesper