netdev - Re: [PATCH net-next 4/5] selftests: hw-net: toeplitz: read indirection table from the device

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <willemdebruijn.kernel.224bdf2fac125@gmail.com>
Date: Fri, 21 Nov 2025 18:12:16 -0500
From: Willem de Bruijn <willemdebruijn.kernel@...il.com>
To: Jakub Kicinski <kuba@...nel.org>, 
 davem@...emloft.net
Cc: netdev@...r.kernel.org, 
 edumazet@...gle.com, 
 pabeni@...hat.com, 
 andrew+netdev@...n.ch, 
 horms@...nel.org, 
 willemb@...gle.com, 
 petrm@...dia.com, 
 dw@...idwei.uk, 
 shuah@...nel.org, 
 linux-kselftest@...r.kernel.org, 
 Jakub Kicinski <kuba@...nel.org>
Subject: Re: [PATCH net-next 4/5] selftests: hw-net: toeplitz: read
 indirection table from the device

Jakub Kicinski wrote:
> Replace the simple modulo math with the real indirection table
> read from the device. This makes the tests pass for mlx5 and
> bnxt NICs.
> 
> Signed-off-by: Jakub Kicinski <kuba@...nel.org>
> ---
>  .../selftests/drivers/net/hw/toeplitz.c       | 24 ++++++++++++++++++-
>  1 file changed, 23 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/testing/selftests/drivers/net/hw/toeplitz.c b/tools/testing/selftests/drivers/net/hw/toeplitz.c
> index 7420a4e201cc..a4d04438c313 100644
> --- a/tools/testing/selftests/drivers/net/hw/toeplitz.c
> +++ b/tools/testing/selftests/drivers/net/hw/toeplitz.c
> @@ -68,6 +68,7 @@
>  #define FOUR_TUPLE_MAX_LEN	((sizeof(struct in6_addr) * 2) + (sizeof(uint16_t) * 2))
>  
>  #define RSS_MAX_CPUS (1 << 16)	/* real constraint is PACKET_FANOUT_MAX */
> +#define RSS_MAX_INDIR	(1 << 16)

Only if respinning, maybe also fix alignment of RSS_MAX_CPUS
  
>  #define RPS_MAX_CPUS 16UL	/* must be a power of 2 */
>  
> @@ -105,6 +106,8 @@ struct ring_state {
>  static unsigned int rx_irq_cpus[RSS_MAX_CPUS];	/* map from rxq to cpu */
>  static int rps_silo_to_cpu[RPS_MAX_CPUS];
>  static unsigned char toeplitz_key[TOEPLITZ_KEY_MAX_LEN];
> +static unsigned int rss_indir_tbl[RSS_MAX_INDIR];
> +static unsigned int rss_indir_tbl_size;
>  static struct ring_state rings[RSS_MAX_CPUS];
>  
>  static inline uint32_t toeplitz(const unsigned char *four_tuple,
> @@ -133,7 +136,12 @@ static inline uint32_t toeplitz(const unsigned char *four_tuple,
>  /* Compare computed cpu with arrival cpu from packet_fanout_cpu */
>  static void verify_rss(uint32_t rx_hash, int cpu)
>  {
> -	int queue = rx_hash % cfg_num_queues;
> +	int queue;
> +
> +	if (rss_indir_tbl_size)
> +		queue = rss_indir_tbl[rx_hash % rss_indir_tbl_size];
> +	else
> +		queue = rx_hash % cfg_num_queues;
>  
>  	log_verbose(" rxq %d (cpu %d)", queue, rx_irq_cpus[queue]);
>  	if (rx_irq_cpus[queue] != cpu) {
> @@ -517,6 +525,20 @@ static void read_rss_dev_info_ynl(void)
>  
>  	memcpy(toeplitz_key, rsp->hkey, rsp->_len.hkey);
>  
> +	if (rsp->_count.indir > RSS_MAX_INDIR)
> +		error(1, 0, "RSS indirection table too large (%u > %u)",
> +		      rsp->_count.indir, RSS_MAX_INDIR);
> +
> +	/* If indir table not available we'll fallback to simple modulo math */
> +	if (rsp->_count.indir) {
> +		memcpy(rss_indir_tbl, rsp->indir,
> +		       rsp->_count.indir * sizeof(rss_indir_tbl[0]));

It can be assumed that rsp->indir elements are sizeof(rss_indir_tbl[0])?

Is there a way to have the test verify element size. I'm not that
familiar with YNL.

> +		rss_indir_tbl_size = rsp->_count.indir;
> +
> +		log_verbose("RSS indirection table size: %u\n",
> +			    rss_indir_tbl_size);
> +	}
> +
>  	ethtool_rss_get_rsp_free(rsp);
>  	ethtool_rss_get_req_free(req);
>  	ynl_sock_destroy(ys);
> -- 
> 2.51.1
>