[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <a66335ec-0911-1902-7585-827b918fe957@nbd.name>
Date: Sun, 19 Feb 2023 16:08:19 +0100
From: Felix Fietkau <nbd@....name>
To: netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>
Cc: linux-kernel@...r.kernel.org
Subject: Re: [RFC v3] net/core: add optional threading for backlog processing
On 19.02.23 14:10, Felix Fietkau wrote:
> When dealing with few flows or an imbalance on CPU utilization, static RPS
> CPU assignment can be too inflexible. Add support for enabling threaded NAPI
> for backlog processing in order to allow the scheduler to better balance
> processing. This helps better spread the load across idle CPUs.
>
> Signed-off-by: Felix Fietkau <nbd@....name>
> ---
> RFC v3:
> - make patch more generic, applies to backlog processing in general
> - fix process queue access on flush
> RFC v2:
> - fix rebase error in rps locking
>
> include/linux/netdevice.h | 2 +
> net/core/dev.c | 78 +++++++++++++++++++++++++++++++++++---
> net/core/sysctl_net_core.c | 27 +++++++++++++
> 3 files changed, 102 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index d9cdbc047b49..b3cef91b1696 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -522,6 +522,7 @@ static inline bool napi_complete(struct napi_struct *n)
> }
>
> int dev_set_threaded(struct net_device *dev, bool threaded);
> +int backlog_set_threaded(bool threaded);
>
> /**
> * napi_disable - prevent NAPI from scheduling
> @@ -3192,6 +3193,7 @@ struct softnet_data {
> unsigned int cpu;
> unsigned int input_queue_tail;
> #endif
> + unsigned int process_queue_empty;
> unsigned int received_rps;
> unsigned int dropped;
> struct sk_buff_head input_pkt_queue;
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 357081b0113c..76874513b7b5 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -4597,7 +4597,7 @@ static int napi_schedule_rps(struct softnet_data *sd)
> struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
>
> #ifdef CONFIG_RPS
> - if (sd != mysd) {
> + if (sd != mysd && !test_bit(NAPI_STATE_THREADED, &sd->backlog.state)) {
> sd->rps_ipi_next = mysd->rps_ipi_list;
> mysd->rps_ipi_list = sd;
>
> @@ -5778,6 +5778,8 @@ static DEFINE_PER_CPU(struct work_struct, flush_works);
> /* Network device is going away, flush any packets still pending */
> static void flush_backlog(struct work_struct *work)
> {
> + unsigned int process_queue_empty;
> + bool threaded, flush_processq;
> struct sk_buff *skb, *tmp;
> struct softnet_data *sd;
>
> @@ -5792,8 +5794,15 @@ static void flush_backlog(struct work_struct *work)
> input_queue_head_incr(sd);
> }
> }
> +
> + threaded = test_bit(NAPI_STATE_THREADED, &sd->backlog.state);
> + flush_processq = threaded &&
> + !skb_queue_empty_lockless(&sd->process_queue);
Sorry, the patch was missing these lines:
if (flush_processq)
process_queue_empty = sd->process_queue_empty;
> rps_unlock_irq_enable(sd);
>
> + if (threaded)
> + goto out;
> +
> skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
> if (skb->dev->reg_state == NETREG_UNREGISTERING) {
> __skb_unlink(skb, &sd->process_queue);
Powered by blists - more mailing lists