[PATCH net-next v4 0/3] implement kthread based napi poll

[PATCH net-next v4 0/3] implement kthread based napi poll

[Wei Wang]

The idea of moving the napi poll process out of softirq context to a
kernel thread based context is not new.
Paolo Abeni and Hannes Frederic Sowa have proposed patches to move napi
poll to kthread back in 2016. And Felix Fietkau has also proposed
patches of similar ideas to use workqueue to process napi poll just a
few weeks ago.

The main reason we'd like to push forward with this idea is that the
scheduler has poor visibility into cpu cycles spent in softirq context,
and is not able to make optimal scheduling decisions of the user threads.
For example, we see in one of the application benchmark where network
load is high, the CPUs handling network softirqs has ~80% cpu util. And
user threads are still scheduled on those CPUs, despite other more idle
cpus available in the system. And we see very high tail latencies. In this
case, we have to explicitly pin away user threads from the CPUs handling
network softirqs to ensure good performance.
With napi poll moved to kthread, scheduler is in charge of scheduling both
the kthreads handling network load, and the user threads, and is able to
make better decisions. In the previous benchmark, if we do this and we
pin the kthreads processing napi poll to specific CPUs, scheduler is
able to schedule user threads away from these CPUs automatically.

And the reason we prefer 1 kthread per napi, instead of 1 workqueue
entity per host, is that kthread is more configurable than workqueue,
and we could leverage existing tuning tools for threads, like taskset,
chrt, etc to tune scheduling class and cpu set, etc. Another reason is
if we eventually want to provide busy poll feature using kernel threads
for napi poll, kthread seems to be more suitable than workqueue.
Furthermore, for large platforms with 2 NICs attached to 2 sockets,
kthread is more flexible to be pinned to different sets of CPUs.  

In this patch series, I revived Paolo and Hannes's patch in 2016 and
made modifications. Then there are changes proposed by Felix, Jakub,
Paolo and myself on top of those, with suggestions from Eric Dumazet.

In terms of performance, I ran tcp_rr tests with 1000 flows with
various request/response sizes, with RFS/RPS disabled, and compared
performance between softirq vs kthread vs workqueue (patchset proposed
by Felix Fietkau).
Host has 56 hyper threads and 100Gbps nic, 8 rx queues and only 1 numa
node. All threads are unpinned.

        req/resp   QPS   50%tile    90%tile    99%tile    99.9%tile
softirq   1B/1B   2.75M   337us       376us      1.04ms     3.69ms
kthread   1B/1B   2.67M   371us       408us      455us      550us
workq     1B/1B   2.56M   384us       435us      673us      822us

softirq 5KB/5KB   1.46M   678us       750us      969us      2.78ms
kthread 5KB/5KB   1.44M   695us       789us      891us      1.06ms
workq   5KB/5KB   1.34M   720us       905us     1.06ms      1.57ms

softirq 1MB/1MB   11.0K   79ms       166ms      306ms       630ms
kthread 1MB/1MB   11.0K   75ms       177ms      303ms       596ms
workq   1MB/1MB   11.0K   79ms       180ms      303ms       587ms

When running workqueue implementation, I found the number of threads
used is usually twice as much as kthread implementation. This probably
introduces higher scheduling cost, which results in higher tail
latencies in most cases.

I also ran an application benchmark, which performs fixed qps remote SSD
read/write operations, with various sizes. Again, both with RFS/RPS
disabled.
The result is as follows:
         op_size  QPS   50%tile 95%tile 99%tile 99.9%tile  
softirq   4K     572.6K   385us   1.5ms  3.16ms   6.41ms
kthread   4K     572.6K   390us   803us  2.21ms   6.83ms
workq     4k     572.6K   384us   763us  3.12ms   6.87ms

softirq   64K    157.9K   736us   1.17ms 3.40ms   13.75ms
kthread   64K    157.9K   745us   1.23ms 2.76ms    9.87ms 
workq     64K    157.9K   746us   1.23ms 2.76ms    9.96ms

softirq   1M     10.98K   2.03ms  3.10ms  3.7ms   11.56ms
kthread   1M     10.98K   2.13ms  3.21ms  4.02ms  13.3ms
workq     1M     10.98K   2.13ms  3.20ms  3.99ms  14.12ms

In this set of tests, the latency is predominant by the SSD operation.
Also, the user threads are much busier compared to tcp_rr tests. We have
to pin the kthreads/workqueue threads to limit to a few CPUs, to not
disturb user threads, and provide some isolation.

Changes since v3:
Merged and rearranged patches in a logical order for easier review. 
Changed sysfs control to be per device.

Changes since v2:
Corrected typo in patch 1, and updated the cover letter with more
detailed and updated test results.

Changes since v1:
Replaced kthread_create() with kthread_run() in patch 5 as suggested by
Felix Fietkau.

Changes since RFC:
Renamed the kthreads to be napi/<dev>-<napi_id> in patch 5 as suggested
by Hannes Frederic Sowa.

Felix Fietkau (1):
  net: extract napi poll functionality to __napi_poll()
Wei Wang (2):
  net: implement threaded-able napi poll loop support
  net: add sysfs attribute to control napi threaded mode

 include/linux/netdevice.h |   5 ++
 net/core/dev.c            | 140 +++++++++++++++++++++++++++++++++++---
 net/core/net-sysfs.c      |  70 +++++++++++++++++++
 3 files changed, 205 insertions(+), 10 deletions(-)

-- 
2.29.2.576.ga3fc446d84-goog

[PATCH net-next v4 1/3] net: extract napi poll functionality to __napi_poll()

[Wei Wang]

From: Felix Fietkau <nbd@nbd.name> 

This commit introduces a new function __napi_poll() which does the main
logic of the existing napi_poll() function, and will be called by other
functions in later commits.
This idea and implementation is done by Felix Fietkau <nbd@nbd.name> and
is proposed as part of the patch to move napi work to work_queue
context.
This commit by itself is a code restructure.

Signed-off-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Wei Wang <weiwan@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
---
 net/core/dev.c | 35 +++++++++++++++++++++++++----------
 1 file changed, 25 insertions(+), 10 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index ce8fea2e2788..8064af1dd03c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -6765,15 +6765,10 @@ void __netif_napi_del(struct napi_struct *napi)
 }
 EXPORT_SYMBOL(__netif_napi_del);
 
-static int napi_poll(struct napi_struct *n, struct list_head *repoll)
+static int __napi_poll(struct napi_struct *n, bool *repoll)
 {
-	void *have;
 	int work, weight;
 
-	list_del_init(&n->poll_list);
-
-	have = netpoll_poll_lock(n);
-
 	weight = n->weight;
 
 	/* This NAPI_STATE_SCHED test is for avoiding a race
@@ -6793,7 +6788,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 			    n->poll, work, weight);
 
 	if (likely(work < weight))
-		goto out_unlock;
+		return work;
 
 	/* Drivers must not modify the NAPI state if they
 	 * consume the entire weight.  In such cases this code
@@ -6802,7 +6797,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 	 */
 	if (unlikely(napi_disable_pending(n))) {
 		napi_complete(n);
-		goto out_unlock;
+		return work;
 	}
 
 	/* The NAPI context has more processing work, but busy-polling
@@ -6815,7 +6810,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 			 */
 			napi_schedule(n);
 		}
-		goto out_unlock;
+		return work;
 	}
 
 	if (n->gro_bitmask) {
@@ -6833,9 +6828,29 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 	if (unlikely(!list_empty(&n->poll_list))) {
 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
 			     n->dev ? n->dev->name : "backlog");
-		goto out_unlock;
+		return work;
 	}
 
+	*repoll = true;
+
+	return work;
+}
+
+static int napi_poll(struct napi_struct *n, struct list_head *repoll)
+{
+	bool do_repoll = false;
+	void *have;
+	int work;
+
+	list_del_init(&n->poll_list);
+
+	have = netpoll_poll_lock(n);
+
+	work = __napi_poll(n, &do_repoll);
+
+	if (!do_repoll)
+		goto out_unlock;
+
 	list_add_tail(&n->poll_list, repoll);
 
 out_unlock:
-- 
2.29.2.576.ga3fc446d84-goog

[PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Wei Wang]

This patch allows running each napi poll loop inside its own
kernel thread.
The threaded mode could be enabled through napi_set_threaded()
api, and does not require a device up/down. The kthread gets
created on demand when napi_set_threaded() is called, and gets
shut down eventually in napi_disable().

Once that threaded mode is enabled and the kthread is
started, napi_schedule() will wake-up such thread instead
of scheduling the softirq.

The threaded poll loop behaves quite likely the net_rx_action,
but it does not have to manipulate local irqs and uses
an explicit scheduling point based on netdev_budget.

Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Co-developed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Wei Wang <weiwan@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h |   5 ++
 net/core/dev.c            | 105 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 110 insertions(+)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 7bf167993c05..abd3b52b7da6 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -347,6 +347,7 @@ struct napi_struct {
 	struct list_head	dev_list;
 	struct hlist_node	napi_hash_node;
 	unsigned int		napi_id;
+	struct task_struct	*thread;
 };
 
 enum {
@@ -358,6 +359,7 @@ enum {
 	NAPI_STATE_NO_BUSY_POLL,	/* Do not add in napi_hash, no busy polling */
 	NAPI_STATE_IN_BUSY_POLL,	/* sk_busy_loop() owns this NAPI */
 	NAPI_STATE_PREFER_BUSY_POLL,	/* prefer busy-polling over softirq processing*/
+	NAPI_STATE_THREADED,		/* The poll is performed inside its own thread*/
 };
 
 enum {
@@ -369,6 +371,7 @@ enum {
 	NAPIF_STATE_NO_BUSY_POLL	= BIT(NAPI_STATE_NO_BUSY_POLL),
 	NAPIF_STATE_IN_BUSY_POLL	= BIT(NAPI_STATE_IN_BUSY_POLL),
 	NAPIF_STATE_PREFER_BUSY_POLL	= BIT(NAPI_STATE_PREFER_BUSY_POLL),
+	NAPIF_STATE_THREADED		= BIT(NAPI_STATE_THREADED),
 };
 
 enum gro_result {
@@ -495,6 +498,8 @@ static inline bool napi_complete(struct napi_struct *n)
 	return napi_complete_done(n, 0);
 }
 
+int napi_set_threaded(struct napi_struct *n, bool threaded);
+
 /**
  *	napi_disable - prevent NAPI from scheduling
  *	@n: NAPI context
diff --git a/net/core/dev.c b/net/core/dev.c
index 8064af1dd03c..cb6c4e2363a4 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -91,6 +91,7 @@
 #include <linux/etherdevice.h>
 #include <linux/ethtool.h>
 #include <linux/skbuff.h>
+#include <linux/kthread.h>
 #include <linux/bpf.h>
 #include <linux/bpf_trace.h>
 #include <net/net_namespace.h>
@@ -1475,6 +1476,36 @@ void netdev_notify_peers(struct net_device *dev)
 }
 EXPORT_SYMBOL(netdev_notify_peers);
 
+static int napi_threaded_poll(void *data);
+
+static int napi_kthread_create(struct napi_struct *n)
+{
+	int err = 0;
+
+	/* Create and wake up the kthread once to put it in
+	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
+	 * warning and work with loadavg.
+	 */
+	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
+				n->dev->name, n->napi_id);
+	if (IS_ERR(n->thread)) {
+		err = PTR_ERR(n->thread);
+		pr_err("kthread_run failed with err %d\n", err);
+		n->thread = NULL;
+	}
+
+	return err;
+}
+
+static void napi_kthread_stop(struct napi_struct *n)
+{
+	if (!n->thread)
+		return;
+	kthread_stop(n->thread);
+	clear_bit(NAPI_STATE_THREADED, &n->state);
+	n->thread = NULL;
+}
+
 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
 {
 	const struct net_device_ops *ops = dev->netdev_ops;
@@ -4234,6 +4265,11 @@ int gro_normal_batch __read_mostly = 8;
 static inline void ____napi_schedule(struct softnet_data *sd,
 				     struct napi_struct *napi)
 {
+	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
+		wake_up_process(napi->thread);
+		return;
+	}
+
 	list_add_tail(&napi->poll_list, &sd->poll_list);
 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
 }
@@ -6690,6 +6726,29 @@ static void init_gro_hash(struct napi_struct *napi)
 	napi->gro_bitmask = 0;
 }
 
+int napi_set_threaded(struct napi_struct *n, bool threaded)
+{
+	int err = 0;
+
+	ASSERT_RTNL();
+
+	if (threaded == !!test_bit(NAPI_STATE_THREADED, &n->state))
+		return 0;
+	if (threaded) {
+		if (!n->thread) {
+			err = napi_kthread_create(n);
+			if (err)
+				goto out;
+		}
+		set_bit(NAPI_STATE_THREADED, &n->state);
+	} else {
+		clear_bit(NAPI_STATE_THREADED, &n->state);
+	}
+
+out:
+	return err;
+}
+
 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 		    int (*poll)(struct napi_struct *, int), int weight)
 {
@@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
 		msleep(1);
 
 	hrtimer_cancel(&n->timer);
+	napi_kthread_stop(n);
 
 	clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
 	clear_bit(NAPI_STATE_DISABLE, &n->state);
@@ -6859,6 +6919,51 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 	return work;
 }
 
+static int napi_thread_wait(struct napi_struct *napi)
+{
+	set_current_state(TASK_INTERRUPTIBLE);
+
+	while (!kthread_should_stop() && !napi_disable_pending(napi)) {
+		if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
+			WARN_ON(!list_empty(&napi->poll_list));
+			__set_current_state(TASK_RUNNING);
+			return 0;
+		}
+
+		schedule();
+		set_current_state(TASK_INTERRUPTIBLE);
+	}
+	__set_current_state(TASK_RUNNING);
+	return -1;
+}
+
+static int napi_threaded_poll(void *data)
+{
+	struct napi_struct *napi = data;
+	void *have;
+
+	while (!napi_thread_wait(napi)) {
+		for (;;) {
+			bool repoll = false;
+
+			local_bh_disable();
+
+			have = netpoll_poll_lock(napi);
+			__napi_poll(napi, &repoll);
+			netpoll_poll_unlock(have);
+
+			__kfree_skb_flush();
+			local_bh_enable();
+
+			if (!repoll)
+				break;
+
+			cond_resched();
+		}
+	}
+	return 0;
+}
+
 static __latent_entropy void net_rx_action(struct softirq_action *h)
 {
 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
-- 
2.29.2.576.ga3fc446d84-goog

[PATCH net-next v4 3/3] net: add sysfs attribute to control napi threaded mode

[Wei Wang]

This patch adds a new sysfs attribute to the network device class.
Said attribute provides a per-device control to enable/disable the
threaded mode for all the napi instances of the given network device.

Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Co-developed-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Wei Wang <weiwan@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
---
 net/core/net-sysfs.c | 70 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 70 insertions(+)

diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index 94fff0700bdd..d14dc1da4c97 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -538,6 +538,75 @@ static ssize_t phys_switch_id_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(phys_switch_id);
 
+static ssize_t threaded_show(struct device *dev,
+			     struct device_attribute *attr, char *buf)
+{
+	struct net_device *netdev = to_net_dev(dev);
+	struct napi_struct *n;
+	bool enabled;
+	int ret;
+
+	if (!rtnl_trylock())
+		return restart_syscall();
+
+	if (!dev_isalive(netdev)) {
+		ret = -EINVAL;
+		goto unlock;
+	}
+
+	if (list_empty(&netdev->napi_list)) {
+		ret = -EOPNOTSUPP;
+		goto unlock;
+	}
+
+	n = list_first_entry(&netdev->napi_list, struct napi_struct, dev_list);
+	enabled = !!test_bit(NAPI_STATE_THREADED, &n->state);
+
+	ret = sprintf(buf, fmt_dec, enabled);
+
+unlock:
+	rtnl_unlock();
+	return ret;
+}
+
+static void dev_disable_threaded_all(struct net_device *dev)
+{
+	struct napi_struct *napi;
+
+	list_for_each_entry(napi, &dev->napi_list, dev_list)
+		napi_set_threaded(napi, false);
+}
+
+static int modify_napi_threaded(struct net_device *dev, unsigned long val)
+{
+	struct napi_struct *napi;
+	int ret;
+
+	if (list_empty(&dev->napi_list))
+		return -EOPNOTSUPP;
+
+	list_for_each_entry(napi, &dev->napi_list, dev_list) {
+		ret = napi_set_threaded(napi, !!val);
+		if (ret) {
+			/* Error occurred on one of the napi,
+			 * reset threaded mode on all napi.
+			 */
+			dev_disable_threaded_all(dev);
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static ssize_t threaded_store(struct device *dev,
+			      struct device_attribute *attr,
+			      const char *buf, size_t len)
+{
+	return netdev_store(dev, attr, buf, len, modify_napi_threaded);
+}
+static DEVICE_ATTR_RW(threaded);
+
 static struct attribute *net_class_attrs[] __ro_after_init = {
 	&dev_attr_netdev_group.attr,
 	&dev_attr_type.attr,
@@ -570,6 +639,7 @@ static struct attribute *net_class_attrs[] __ro_after_init = {
 	&dev_attr_proto_down.attr,
 	&dev_attr_carrier_up_count.attr,
 	&dev_attr_carrier_down_count.attr,
+	&dev_attr_threaded.attr,
 	NULL,
 };
 ATTRIBUTE_GROUPS(net_class);
-- 
2.29.2.576.ga3fc446d84-goog

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Jakub Kicinski]

On Tue,  8 Dec 2020 16:54:43 -0800 Wei Wang wrote:
> This patch allows running each napi poll loop inside its own
> kernel thread.
> The threaded mode could be enabled through napi_set_threaded()
> api, and does not require a device up/down. The kthread gets
> created on demand when napi_set_threaded() is called, and gets
> shut down eventually in napi_disable().
> 
> Once that threaded mode is enabled and the kthread is
> started, napi_schedule() will wake-up such thread instead
> of scheduling the softirq.
> 
> The threaded poll loop behaves quite likely the net_rx_action,
> but it does not have to manipulate local irqs and uses
> an explicit scheduling point based on netdev_budget.
> 
> Co-developed-by: Paolo Abeni <pabeni@redhat.com>
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> Co-developed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
> Co-developed-by: Jakub Kicinski <kuba@kernel.org>
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> Signed-off-by: Wei Wang <weiwan@google.com>
> Reviewed-by: Eric Dumazet <edumazet@google.com>

> @@ -4234,6 +4265,11 @@ int gro_normal_batch __read_mostly = 8;
>  static inline void ____napi_schedule(struct softnet_data *sd,
>  				     struct napi_struct *napi)
>  {
> +	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
> +		wake_up_process(napi->thread);

FTR your implementation depends on the fact that this is the only
place that can wake the worker and not set kthread_should_stop().
Which I trust you is the case :) maybe I already mentioned this..

> +		return;
> +	}
> +
>  	list_add_tail(&napi->poll_list, &sd->poll_list);
>  	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
>  }

>  void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
>  		    int (*poll)(struct napi_struct *, int), int weight)
>  {
> @@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
>  		msleep(1);
>  
>  	hrtimer_cancel(&n->timer);
> +	napi_kthread_stop(n);

I'm surprised that we stop the thread on napi_disable() but there is no
start/create in napi_enable(). NAPIs can (and do get) disabled and
enabled again. But that'd make your code crash with many popular
drivers if you tried to change rings with threaded napi enabled so I
feel like I must be missing something..

>  	clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
>  	clear_bit(NAPI_STATE_DISABLE, &n->state);

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Jakub Kicinski]

On Sat, 12 Dec 2020 14:50:22 -0800 Jakub Kicinski wrote:
> > @@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
> >  		msleep(1);
> >  
> >  	hrtimer_cancel(&n->timer);
> > +	napi_kthread_stop(n);  
> 
> I'm surprised that we stop the thread on napi_disable() but there is no
> start/create in napi_enable(). NAPIs can (and do get) disabled and
> enabled again. But that'd make your code crash with many popular
> drivers if you tried to change rings with threaded napi enabled so I
> feel like I must be missing something..

Ah, not crash, 'cause the flag gets cleared. Is it intentional that any
changes that disable NAPIs cause us to go back to non-threaded NAPI?
I think I had the "threaded" setting stored in struct netdevice in my
patches, is there a reason not to do that?

In fact your patches may _require_ the device to be up to enable
threaded NAPI if NAPIs are allocated in open.

Re: [PATCH net-next v4 3/3] net: add sysfs attribute to control napi threaded mode

[Jakub Kicinski]

On Tue,  8 Dec 2020 16:54:44 -0800 Wei Wang wrote:
> +static void dev_disable_threaded_all(struct net_device *dev)
> +{
> +	struct napi_struct *napi;
> +
> +	list_for_each_entry(napi, &dev->napi_list, dev_list)
> +		napi_set_threaded(napi, false);
> +}

This is an implementation detail which should be hidden in dev.c IMHO.
Since the sysfs knob is just a device global on/off the iteration over
napis and all is best hidden away.

(sorry about the delayed review BTW, hope we can do a minor revision
and still hit 5.12)

> +static int modify_napi_threaded(struct net_device *dev, unsigned long val)
> +{
> +	struct napi_struct *napi;
> +	int ret;
> +
> +	if (list_empty(&dev->napi_list))
> +		return -EOPNOTSUPP;
> +
> +	list_for_each_entry(napi, &dev->napi_list, dev_list) {
> +		ret = napi_set_threaded(napi, !!val);
> +		if (ret) {
> +			/* Error occurred on one of the napi,
> +			 * reset threaded mode on all napi.
> +			 */
> +			dev_disable_threaded_all(dev);
> +			break;
> +		}
> +	}
> +
> +	return ret;
> +}

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Wei Wang]

On Sat, Dec 12, 2020 at 2:55 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Sat, 12 Dec 2020 14:50:22 -0800 Jakub Kicinski wrote:
> > > @@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
> > >             msleep(1);
> > >
> > >     hrtimer_cancel(&n->timer);
> > > +   napi_kthread_stop(n);
> >
> > I'm surprised that we stop the thread on napi_disable() but there is no
> > start/create in napi_enable(). NAPIs can (and do get) disabled and
> > enabled again. But that'd make your code crash with many popular
> > drivers if you tried to change rings with threaded napi enabled so I
> > feel like I must be missing something..
>
> Ah, not crash, 'cause the flag gets cleared. Is it intentional that any
> changes that disable NAPIs cause us to go back to non-threaded NAPI?
> I think I had the "threaded" setting stored in struct netdevice in my
> patches, is there a reason not to do that?
>

Thanks for the comments!

The reason that I did not record it in dev is: there is a slight
chance that during creation of the kthreads, failures occur and we
flip back all NAPIs to use non-threaded mode. I am not sure the
recorded value in dev should be what user desires, or what the actual
situation is. Same as after the driver does a
napi_disabe()/napi_enable(). It might occur that the dev->threaded =
true, but the operation to re-create the kthreads fail and we flip
back to non-thread mode. This seems to get things more complicated.
What I expect is the user only enables the threaded mode after the
device is up and alive, with all NAPIs attached to dev, and enabled.
And user has to check the sysfs to make sure that the operation
succeeds.
And any operation that brings down the device, will flip this back to
default, which is non-threaded mode.

> In fact your patches may _require_ the device to be up to enable
> threaded NAPI if NAPIs are allocated in open.

Re: [PATCH net-next v4 3/3] net: add sysfs attribute to control napi threaded mode

[Wei Wang]

On Sat, Dec 12, 2020 at 2:59 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Tue,  8 Dec 2020 16:54:44 -0800 Wei Wang wrote:
> > +static void dev_disable_threaded_all(struct net_device *dev)
> > +{
> > +     struct napi_struct *napi;
> > +
> > +     list_for_each_entry(napi, &dev->napi_list, dev_list)
> > +             napi_set_threaded(napi, false);
> > +}
>
> This is an implementation detail which should be hidden in dev.c IMHO.
> Since the sysfs knob is just a device global on/off the iteration over
> napis and all is best hidden away.

OK. Will move it.

>
> (sorry about the delayed review BTW, hope we can do a minor revision
> and still hit 5.12)
>
> > +static int modify_napi_threaded(struct net_device *dev, unsigned long val)
> > +{
> > +     struct napi_struct *napi;
> > +     int ret;
> > +
> > +     if (list_empty(&dev->napi_list))
> > +             return -EOPNOTSUPP;
> > +
> > +     list_for_each_entry(napi, &dev->napi_list, dev_list) {
> > +             ret = napi_set_threaded(napi, !!val);
> > +             if (ret) {
> > +                     /* Error occurred on one of the napi,
> > +                      * reset threaded mode on all napi.
> > +                      */
> > +                     dev_disable_threaded_all(dev);
> > +                     break;
> > +             }
> > +     }
> > +
> > +     return ret;
> > +}

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Jakub Kicinski]

On Mon, 14 Dec 2020 09:59:21 -0800 Wei Wang wrote:
> On Sat, Dec 12, 2020 at 2:55 PM Jakub Kicinski <kuba@kernel.org> wrote:
> >
> > On Sat, 12 Dec 2020 14:50:22 -0800 Jakub Kicinski wrote:  
> > > > @@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
> > > >             msleep(1);
> > > >
> > > >     hrtimer_cancel(&n->timer);
> > > > +   napi_kthread_stop(n);  
> > >
> > > I'm surprised that we stop the thread on napi_disable() but there is no
> > > start/create in napi_enable(). NAPIs can (and do get) disabled and
> > > enabled again. But that'd make your code crash with many popular
> > > drivers if you tried to change rings with threaded napi enabled so I
> > > feel like I must be missing something..  
> >
> > Ah, not crash, 'cause the flag gets cleared. Is it intentional that any
> > changes that disable NAPIs cause us to go back to non-threaded NAPI?
> > I think I had the "threaded" setting stored in struct netdevice in my
> > patches, is there a reason not to do that?
> 
> Thanks for the comments!
> 
> The reason that I did not record it in dev is: there is a slight
> chance that during creation of the kthreads, failures occur and we
> flip back all NAPIs to use non-threaded mode. I am not sure the
> recorded value in dev should be what user desires, or what the actual
> situation is. Same as after the driver does a
> napi_disabe()/napi_enable(). It might occur that the dev->threaded =
> true, but the operation to re-create the kthreads fail and we flip
> back to non-thread mode. This seems to get things more complicated.
> What I expect is the user only enables the threaded mode after the
> device is up and alive, with all NAPIs attached to dev, and enabled.
> And user has to check the sysfs to make sure that the operation
> succeeds.
> And any operation that brings down the device, will flip this back to
> default, which is non-threaded mode.

It is quite an annoying problem to address, given all relevant NAPI
helpers seem to return void :/ But we're pushing the problem onto the
user just because of internal API structure.

This reminds me of PTP / timestamping issues some NICs had once upon 
a time. The timing application enables HW time stamping, then later some
other application / orchestration changes a seemingly unrelated config,
and since NIC has to reset itself it looses the timestamping config.
Now the time app stops getting HW time stamps, but those are best
effort anyway, so it just assumes the NIC couldn't stamp given frame
(for every frame), not that config got completely broken. The system
keeps running with suboptimal time for months.

What does the deployment you're expecting to see looks like? What
entity controls enabling the threaded mode on a system? Application?
Orchestration? What's the flow?

"Forgetting" config based on driver-dependent events feels very fragile.

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Wei Wang]

On Mon, Dec 14, 2020 at 11:02 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Mon, 14 Dec 2020 09:59:21 -0800 Wei Wang wrote:
> > On Sat, Dec 12, 2020 at 2:55 PM Jakub Kicinski <kuba@kernel.org> wrote:
> > >
> > > On Sat, 12 Dec 2020 14:50:22 -0800 Jakub Kicinski wrote:
> > > > > @@ -6731,6 +6790,7 @@ void napi_disable(struct napi_struct *n)
> > > > >             msleep(1);
> > > > >
> > > > >     hrtimer_cancel(&n->timer);
> > > > > +   napi_kthread_stop(n);
> > > >
> > > > I'm surprised that we stop the thread on napi_disable() but there is no
> > > > start/create in napi_enable(). NAPIs can (and do get) disabled and
> > > > enabled again. But that'd make your code crash with many popular
> > > > drivers if you tried to change rings with threaded napi enabled so I
> > > > feel like I must be missing something..
> > >
> > > Ah, not crash, 'cause the flag gets cleared. Is it intentional that any
> > > changes that disable NAPIs cause us to go back to non-threaded NAPI?
> > > I think I had the "threaded" setting stored in struct netdevice in my
> > > patches, is there a reason not to do that?
> >
> > Thanks for the comments!
> >
> > The reason that I did not record it in dev is: there is a slight
> > chance that during creation of the kthreads, failures occur and we
> > flip back all NAPIs to use non-threaded mode. I am not sure the
> > recorded value in dev should be what user desires, or what the actual
> > situation is. Same as after the driver does a
> > napi_disabe()/napi_enable(). It might occur that the dev->threaded =
> > true, but the operation to re-create the kthreads fail and we flip
> > back to non-thread mode. This seems to get things more complicated.
> > What I expect is the user only enables the threaded mode after the
> > device is up and alive, with all NAPIs attached to dev, and enabled.
> > And user has to check the sysfs to make sure that the operation
> > succeeds.
> > And any operation that brings down the device, will flip this back to
> > default, which is non-threaded mode.
>
> It is quite an annoying problem to address, given all relevant NAPI
> helpers seem to return void :/ But we're pushing the problem onto the
> user just because of internal API structure.
>
> This reminds me of PTP / timestamping issues some NICs had once upon
> a time. The timing application enables HW time stamping, then later some
> other application / orchestration changes a seemingly unrelated config,
> and since NIC has to reset itself it looses the timestamping config.
> Now the time app stops getting HW time stamps, but those are best
> effort anyway, so it just assumes the NIC couldn't stamp given frame
> (for every frame), not that config got completely broken. The system
> keeps running with suboptimal time for months.
>
> What does the deployment you're expecting to see looks like? What
> entity controls enabling the threaded mode on a system? Application?
> Orchestration? What's the flow?
>
I see your point. In our deployment, we have a system daemon which is
responsible for setting up all the system tunings after the host boots
up (before application starts to run). If certain operation fails, it
prints out error msg, and will exit with error. For applications that
require threaded mode, I think a check to the sysfs entry to make sure
it is enabled is necessary at the startup phase.


> "Forgetting" config based on driver-dependent events feels very fragile.
I think we could add a recorded value in dev to represent the user
setting, and try to enable threaded mode after napi_disable/enable.
But I think user/application still has to check the sysfs entry value
to make sure if it is enabled successfully.

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Jakub Kicinski]

On Mon, 14 Dec 2020 11:45:43 -0800 Wei Wang wrote:
> > It is quite an annoying problem to address, given all relevant NAPI
> > helpers seem to return void :/ But we're pushing the problem onto the
> > user just because of internal API structure.
> >
> > This reminds me of PTP / timestamping issues some NICs had once upon
> > a time. The timing application enables HW time stamping, then later some
> > other application / orchestration changes a seemingly unrelated config,
> > and since NIC has to reset itself it looses the timestamping config.
> > Now the time app stops getting HW time stamps, but those are best
> > effort anyway, so it just assumes the NIC couldn't stamp given frame
> > (for every frame), not that config got completely broken. The system
> > keeps running with suboptimal time for months.
> >
> > What does the deployment you're expecting to see looks like? What
> > entity controls enabling the threaded mode on a system? Application?
> > Orchestration? What's the flow?
> >  
> I see your point. In our deployment, we have a system daemon which is
> responsible for setting up all the system tunings after the host boots
> up (before application starts to run). If certain operation fails, it
> prints out error msg, and will exit with error. For applications that
> require threaded mode, I think a check to the sysfs entry to make sure
> it is enabled is necessary at the startup phase.

That assumes no workload stacking, and dynamic changes after the
workload has started? Or does the daemon have enough clever logic
to resolve config changes?

> > "Forgetting" config based on driver-dependent events feels very fragile.  
> I think we could add a recorded value in dev to represent the user
> setting, and try to enable threaded mode after napi_disable/enable.
> But I think user/application still has to check the sysfs entry value
> to make sure if it is enabled successfully.

In case of an error you're thinking of resetting, still, and returning
disabled from sysfs? I guess that's fine, we can leave failing the bad
reconfig operation (rather than resetting config) as a future extension.
Let's add a WARN_ON, tho, so the failures don't get missed.

Re: [PATCH net-next v4 2/3] net: implement threaded-able napi poll loop support

[Wei Wang]

On Mon, Dec 14, 2020 at 12:33 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Mon, 14 Dec 2020 11:45:43 -0800 Wei Wang wrote:
> > > It is quite an annoying problem to address, given all relevant NAPI
> > > helpers seem to return void :/ But we're pushing the problem onto the
> > > user just because of internal API structure.
> > >
> > > This reminds me of PTP / timestamping issues some NICs had once upon
> > > a time. The timing application enables HW time stamping, then later some
> > > other application / orchestration changes a seemingly unrelated config,
> > > and since NIC has to reset itself it looses the timestamping config.
> > > Now the time app stops getting HW time stamps, but those are best
> > > effort anyway, so it just assumes the NIC couldn't stamp given frame
> > > (for every frame), not that config got completely broken. The system
> > > keeps running with suboptimal time for months.
> > >
> > > What does the deployment you're expecting to see looks like? What
> > > entity controls enabling the threaded mode on a system? Application?
> > > Orchestration? What's the flow?
> > >
> > I see your point. In our deployment, we have a system daemon which is
> > responsible for setting up all the system tunings after the host boots
> > up (before application starts to run). If certain operation fails, it
> > prints out error msg, and will exit with error. For applications that
> > require threaded mode, I think a check to the sysfs entry to make sure
> > it is enabled is necessary at the startup phase.
>
> That assumes no workload stacking, and dynamic changes after the
> workload has started? Or does the daemon have enough clever logic
> to resolve config changes?
>

The former. There should not be any dynamic changes after the workload
has started. At least for now.

> > > "Forgetting" config based on driver-dependent events feels very fragile.
> > I think we could add a recorded value in dev to represent the user
> > setting, and try to enable threaded mode after napi_disable/enable.
> > But I think user/application still has to check the sysfs entry value
> > to make sure if it is enabled successfully.
>
> In case of an error you're thinking of resetting, still, and returning
> disabled from sysfs? I guess that's fine, we can leave failing the bad
> reconfig operation (rather than resetting config) as a future extension.
> Let's add a WARN_ON, tho, so the failures don't get missed.

Yes. In terms of error, all napi will be reset to using non-threaded
mode, and sysfs returns disabled.
OK. Will add a WARN_ON.