Re: [PATCH net-next] net: introduce budget_squeeze to help us tune rx behavior

[Kui-Feng Lee]

On 3/11/23 08:36, Jason Xing wrote:
> From: Jason Xing <kernelxing@xxxxxxxxxxx>
>
> When we encounter some performance issue and then get lost on how
> to tune the budget limit and time limit in net_rx_action() function,
> we can separately counting both of them to avoid the confusion.
>
> Signed-off-by: Jason Xing <kernelxing@xxxxxxxxxxx>
> ---
> note: this commit is based on the link as below:
> https://lore.kernel.org/lkml/20230311151756.83302-1-kerneljasonxing@xxxxxxxxx/
> ---
>   include/linux/netdevice.h |  1 +
>   net/core/dev.c            | 12 ++++++++----
>   net/core/net-procfs.c     |  9 ++++++---
>   3 files changed, 15 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index 6a14b7b11766..5736311a2133 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -3157,6 +3157,7 @@ struct softnet_data {
>   	/* stats */
>   	unsigned int		processed;
>   	unsigned int		time_squeeze;
> +	unsigned int		budget_squeeze;
>   #ifdef CONFIG_RPS
>   	struct softnet_data	*rps_ipi_list;
>   #endif
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 253584777101..bed7a68fdb5d 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -6637,6 +6637,7 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
>   	unsigned long time_limit = jiffies +
>   		usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
>   	int budget = READ_ONCE(netdev_budget);
> +	bool is_continue = true;
>   	LIST_HEAD(list);
>   	LIST_HEAD(repoll);
>   
> @@ -6644,7 +6645,7 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
>   	list_splice_init(&sd->poll_list, &list);
>   	local_irq_enable();
>   
> -	for (;;) {
> +	for (; is_continue;) {
>   		struct napi_struct *n;
>   
>   		skb_defer_free_flush(sd);
> @@ -6662,10 +6663,13 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
>   		 * Allow this to run for 2 jiffies since which will allow
>   		 * an average latency of 1.5/HZ.
>   		 */
> -		if (unlikely(budget <= 0 ||
> -			     time_after_eq(jiffies, time_limit))) {
> +		if (unlikely(budget <= 0)) {
> +			sd->budget_squeeze++;
> +			is_continue = false;
> +		}
> +		if (unlikely(time_after_eq(jiffies, time_limit))) {
>   			sd->time_squeeze++;
> -			break;
> +			is_continue = false;
>   		}
>   	}
>   
> diff --git a/net/core/net-procfs.c b/net/core/net-procfs.c
> index 97a304e1957a..4d1a499d7c43 100644
> --- a/net/core/net-procfs.c
> +++ b/net/core/net-procfs.c
> @@ -174,14 +174,17 @@ static int softnet_seq_show(struct seq_file *seq, void *v)
>   	 */
>   	seq_printf(seq,
>   		   "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x "
> -		   "%08x %08x\n",
> -		   sd->processed, sd->dropped, sd->time_squeeze, 0,
> +		   "%08x %08x %08x %08x\n",
> +		   sd->processed, sd->dropped,
> +		   0, /* was old way to count time squeeze */

Should we show a proximate number?  For example,
sd->time_squeeze + sd->bud_squeeze.


> +		   0,
>   		   0, 0, 0, 0, /* was fastroute */
>   		   0,	/* was cpu_collision */
>   		   sd->received_rps, flow_limit_count,
>   		   0,	/* was len of two backlog queues */
>   		   (int)seq->index,
> -		   softnet_input_pkt_queue_len(sd), softnet_process_queue_len(sd));
> +		   softnet_input_pkt_queue_len(sd), softnet_process_queue_len(sd),
> +		   sd->time_squeeze, sd->budget_squeeze);
>   	return 0;
>   }
>