Skip to contentSkip to conversation

net: add support for threaded NAPI polling

PATCH In review Felix Fietkau 4 participants 10 messages Jul 29, 2020 – Aug 4, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: add support for threaded NAPI polling

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

 include/linux/netdevice.h |  23 ++++++
 net/core/dev.c            | 162 ++++++++++++++++++++++++++------------
 2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +23 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
 	struct list_head	dev_list;
 	struct hlist_node	napi_hash_node;
 	unsigned int		napi_id;
+	struct work_struct	work;
 };
 
 enum {
@@ -357,6 +358,7 @@ enum {
 	NAPI_STATE_HASHED,	/* In NAPI hash (busy polling possible) */
 	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_THREADED,	/* Use threaded NAPI */
 };
 
 enum {
@@ -367,6 +369,7 @@ enum {
 	NAPIF_STATE_HASHED	 = BIT(NAPI_STATE_HASHED),
 	NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
 	NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+	NAPIF_STATE_THREADED	 = BIT(NAPI_STATE_THREADED),
 };
 
 enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 		    int (*poll)(struct napi_struct *, int), int weight);
 
+/**
+ *	netif_threaded_napi_add - initialize a NAPI context
+ *	@dev:  network device
+ *	@napi: NAPI context
+ *	@poll: polling function
+ *	@weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */
+static inline void netif_threaded_napi_add(struct net_device *dev,
+					   struct napi_struct *napi,
+					   int (*poll)(struct napi_struct *, int),
+					   int weight)
+{
+	set_bit(NAPI_STATE_THREADED, &napi->state);
+	netif_napi_add(dev, napi, poll, weight);
+}
+
 /**
  *	netif_tx_napi_add - initialize a NAPI context
  *	@dev:  network device
net/core/dev.c +110 -52

211 lines not on this page · raw message

Signature · 1 line
2.24.0
#1 lore raw
Quoted 101 lines · Felix Fietkau

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

 include/linux/netdevice.h |  23 ++++++
 net/core/dev.c            | 162 ++++++++++++++++++++++++++------------
 2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +23 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
 	struct list_head	dev_list;
 	struct hlist_node	napi_hash_node;
 	unsigned int		napi_id;
+	struct work_struct	work;
 };
 
 enum {
@@ -357,6 +358,7 @@ enum {
 	NAPI_STATE_HASHED,	/* In NAPI hash (busy polling possible) */
 	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_THREADED,	/* Use threaded NAPI */
 };
 
 enum {
@@ -367,6 +369,7 @@ enum {
 	NAPIF_STATE_HASHED	 = BIT(NAPI_STATE_HASHED),
 	NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
 	NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+	NAPIF_STATE_THREADED	 = BIT(NAPI_STATE_THREADED),
 };
 
 enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 		    int (*poll)(struct napi_struct *, int), int weight);
 
+/**
+ *	netif_threaded_napi_add - initialize a NAPI context
+ *	@dev:  network device
+ *	@napi: NAPI context
+ *	@poll: polling function
+ *	@weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */
+static inline void netif_threaded_napi_add(struct net_device *dev,
+					   struct napi_struct *napi,
+					   int (*poll)(struct napi_struct *, int),
+					   int weight)
+{
+	set_bit(NAPI_STATE_THREADED, &napi->state);
+	netif_napi_add(dev, napi, poll, weight);
+}
+
 /**
  *	netif_tx_napi_add - initialize a NAPI context
  *	@dev:  network device
net/core/dev.c +1 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 struct list_head ptype_all __read_mostly;	/* Taps */
 static struct list_head offload_base __read_mostly;
+static struct workqueue_struct *napi_workq __read_mostly;
 
 static int netif_rx_internal(struct sk_buff *skb);
 static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6294

Quoted 8 lines
net/core/dev.c +5 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
 {
 	unsigned long flags;
 
+	if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+		queue_work(napi_workq, &n->work);
+		return;
+	}
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Just hoping the thread will eventually run seems optimistic to me.

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

#2 lore raw

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 114 lines · Felix Fietkau
Quoted 101 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

  include/linux/netdevice.h |  23 ++++++
  net/core/dev.c            | 162 ++++++++++++++++++++++++++------------
  2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +23 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
  	struct list_head	dev_list;
  	struct hlist_node	napi_hash_node;
  	unsigned int		napi_id;
+	struct work_struct	work;
  };
 
  enum {
@@ -357,6 +358,7 @@ enum {
  	NAPI_STATE_HASHED,	/* In NAPI hash (busy polling possible) */
  	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_THREADED,	/* Use threaded NAPI */
  };
 
  enum {
@@ -367,6 +369,7 @@ enum {
  	NAPIF_STATE_HASHED	 = BIT(NAPI_STATE_HASHED),
  	NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
  	NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+	NAPIF_STATE_THREADED	 = BIT(NAPI_STATE_THREADED),
  };
 
  enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
  void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
  		    int (*poll)(struct napi_struct *, int), int weight);
 
+/**
+ *	netif_threaded_napi_add - initialize a NAPI context
+ *	@dev:  network device
+ *	@napi: NAPI context
+ *	@poll: polling function
+ *	@weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */
+static inline void netif_threaded_napi_add(struct net_device *dev,
+					   struct napi_struct *napi,
+					   int (*poll)(struct napi_struct *, int),
+					   int weight)
+{
+	set_bit(NAPI_STATE_THREADED, &napi->state);
+	netif_napi_add(dev, napi, poll, weight);
+}
+
  /**
   *	netif_tx_napi_add - initialize a NAPI context
   *	@dev:  network device
net/core/dev.c +1 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
  struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
  struct list_head ptype_all __read_mostly;	/* Taps */
  static struct list_head offload_base __read_mostly;
+static struct workqueue_struct *napi_workq __read_mostly;
 
  static int netif_rx_internal(struct sk_buff *skb);
  static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6294

Quoted 8 lines
net/core/dev.c +5 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
  {
  	unsigned long flags;
 
+	if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+		queue_work(napi_workq, &n->work);
+		return;
+	}
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Quoted 6 lines

Just hoping the thread will eventually run seems optimistic to me.

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

Quoted 3 lines
#3 lore raw

From: Sebastian Gottschall

Quoted 1 line

Sent: 30 July 2020 15:30

...

Quoted 10 lines
Quoted 3 lines

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options.
so changing it would just be possible if a special mac80211 based control would be implemented for these drivers.
for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

ISTM that a global flag that made all NAPI callbacks be made from a worker thread rather than softint would be more approriate. Or even something that made the softint callbacks themselves only run an a specific high(ish) priority kernel thread.

While it might slow down setups that need very low ethernet latency it will help those that don't want application RT threads to be 'stolen' by the softint code while they hold application mutex or are waiting to be woken by a cv.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#4 lore raw

Am 30.07.2020 um 17:42 schrieb David Laight:

Quoted 17 lines

From: Sebastian Gottschall

Quoted 1 line

Sent: 30 July 2020 15:30

...

Quoted 9 lines
Quoted 3 lines

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options.
so changing it would just be possible if a special mac80211 based control would be implemented for these drivers.
for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

ISTM that a global flag that made all NAPI callbacks be made from a worker thread rather than softint would be more approriate. Or even something that made the softint callbacks themselves only run an a specific high(ish) priority kernel thread.

Quoted 4 lines

While it might slow down setups that need very low ethernet latency it will help those that don't want application RT threads to be 'stolen' by the softint code while they hold application mutex or are waiting to be woken by a cv.

this will not work either. i already identified drivers which are incompatible with that approach. (marvell mvneta for instance) so threading should be a feature which should be enabled or disabled on known working chipsets. but it cannot be used as set and forget feature for everything unless the specific driver are fixed.

Quoted 6 lines

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#5 lore raw
In reply to Sebastian Gottschall
Quoted 113 lines · Sebastian Gottschall

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 111 lines · Felix Fietkau
Quoted 96 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

  include/linux/netdevice.h |  23 ++++++
  net/core/dev.c            | 162 ++++++++++++++++++++++++++------------

  2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +1 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
       struct list_head    dev_list;
       struct hlist_node    napi_hash_node;
       unsigned int        napi_id;
+    struct work_struct    work;
   };
     enum {
@@ -357,6 +358,7 @@ enum {

      NAPI_STATE_HASHED,    /* In NAPI hash (busy polling possible) */ 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_THREADED,    /* Use threaded NAPI */
  };
    enum {

include/linux/netdevice.h +1 -0
@@ -367,6 +369,7 @@ enum {
       NAPIF_STATE_HASHED     = BIT(NAPI_STATE_HASHED),
       NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
       NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+    NAPIF_STATE_THREADED     = BIT(NAPI_STATE_THREADED),
   };
     enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
   void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
               int (*poll)(struct napi_struct *, int), int weight);

  +/**
+ *    netif_threaded_napi_add - initialize a NAPI context
+ *    @dev:  network device
+ *    @napi: NAPI context
+ *    @poll: polling function
+ *    @weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */

include/linux/netdevice.h +9 -0
+static inline void netif_threaded_napi_add(struct net_device *dev,
+                       struct napi_struct *napi,
+                       int (*poll)(struct napi_struct *, int),
+                       int weight)
+{
+    set_bit(NAPI_STATE_THREADED, &napi->state);
+    netif_napi_add(dev, napi, poll, weight);
+}
+

  /**
   *    netif_tx_napi_add - initialize a NAPI context *    @dev:  network device

net/core/dev.c +0 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
   struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;

  struct list_head ptype_all __read_mostly;    /* Taps */ static struct list_head offload_base __read_mostly; +static struct workqueue_struct *napi_workq __read_mostly; static int netif_rx_internal(struct sk_buff *skb); static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6293

Quoted 7 lines
net/core/dev.c +4 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
   {
       unsigned long flags;
   +    if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+        queue_work(napi_workq, &n->work);
+        return;
+    }
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Just hoping the thread will eventually run seems optimistic to me.

Quoted 6 lines
Quoted 3 lines

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

We do not want to add code in fast path only for one device. We need something truly generic.

I am not saying only the admin can chose, it is fine if a driver does not give the choice and will simply call netif_threaded_napi_add()

#6 lore raw

Am 30.07.2020 um 18:08 schrieb Eric Dumazet:

Quoted 115 lines · Sebastian Gottschall
Quoted 113 lines

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 112 lines · Felix Fietkau
Quoted 96 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

   include/linux/netdevice.h |  23 ++++++
   net/core/dev.c            | 162 ++++++++++++++++++++++++++------------
   2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +22 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
       struct list_head    dev_list;
       struct hlist_node    napi_hash_node;
       unsigned int        napi_id;
+    struct work_struct    work;
   };
     enum {
@@ -357,6 +358,7 @@ enum {
       NAPI_STATE_HASHED,    /* In NAPI hash (busy polling possible) */
       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_THREADED,    /* Use threaded NAPI */
   };
     enum {
@@ -367,6 +369,7 @@ enum {
       NAPIF_STATE_HASHED     = BIT(NAPI_STATE_HASHED),
       NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
       NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+    NAPIF_STATE_THREADED     = BIT(NAPI_STATE_THREADED),
   };
     enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
   void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
               int (*poll)(struct napi_struct *, int), int weight);
   +/**
+ *    netif_threaded_napi_add - initialize a NAPI context
+ *    @dev:  network device
+ *    @napi: NAPI context
+ *    @poll: polling function
+ *    @weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */
+static inline void netif_threaded_napi_add(struct net_device *dev,
+                       struct napi_struct *napi,
+                       int (*poll)(struct napi_struct *, int),
+                       int weight)
+{
+    set_bit(NAPI_STATE_THREADED, &napi->state);
+    netif_napi_add(dev, napi, poll, weight);
+}
+
   /**
    *    netif_tx_napi_add - initialize a NAPI context
    *    @dev:  network device
net/core/dev.c +1 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
   struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
   struct list_head ptype_all __read_mostly;    /* Taps */
   static struct list_head offload_base __read_mostly;
+static struct workqueue_struct *napi_workq __read_mostly;
     static int netif_rx_internal(struct sk_buff *skb);
   static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6293

Quoted 7 lines
net/core/dev.c +4 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
   {
       unsigned long flags;
   +    if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+        queue_work(napi_workq, &n->work);
+        return;
+    }
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Just hoping the thread will eventually run seems optimistic to me.

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

Quoted 7 lines
Quoted 3 lines

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

We do not want to add code in fast path only for one device. We need something truly generic.

I am not saying only the admin can chose, it is fine if a driver does not give the choice and will simply call netif_threaded_napi_add()

what could make sense if the feature can be disabled / enabled, but it will only affect drivers using the netif_threaded_napi_add call, but it should not affect drivers using the old api in any way since not all drivers will work with this feature.

#7 lore raw
Quoted 120 lines · Sebastian Gottschall

Am 30.07.2020 um 18:08 schrieb Eric Dumazet:

Quoted 118 lines · Sebastian Gottschall
Quoted 116 lines

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 106 lines · Felix Fietkau
Quoted 96 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

   include/linux/netdevice.h |  23 ++++++
   net/core/dev.c            | 162 ++++++++++++++++++++++++++------------

   2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +1 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
        struct list_head    dev_list;
        struct hlist_node    napi_hash_node;
        unsigned int        napi_id;
+    struct work_struct    work;
    };
      enum {
@@ -357,6 +358,7 @@ enum {

       NAPI_STATE_HASHED,    /* In NAPI hash (busy polling possible) */ 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_THREADED,    /* Use threaded NAPI */
   };
     enum {

include/linux/netdevice.h +1 -0
@@ -367,6 +369,7 @@ enum {
        NAPIF_STATE_HASHED     = BIT(NAPI_STATE_HASHED),
        NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
        NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+    NAPIF_STATE_THREADED     = BIT(NAPI_STATE_THREADED),
    };
      enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
    void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
                int (*poll)(struct napi_struct *, int), int weight);

   +/**
+ *    netif_threaded_napi_add - initialize a NAPI context
+ *    @dev:  network device
+ *    @napi: NAPI context
+ *    @poll: polling function
+ *    @weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */

include/linux/netdevice.h +9 -0
+static inline void netif_threaded_napi_add(struct net_device *dev,
+                       struct napi_struct *napi,
+                       int (*poll)(struct napi_struct *, int),
+                       int weight)
+{
+    set_bit(NAPI_STATE_THREADED, &napi->state);
+    netif_napi_add(dev, napi, poll, weight);
+}
+

   /**
    *    netif_tx_napi_add - initialize a NAPI context *    @dev:  network device

net/core/dev.c +0 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
    struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;

   struct list_head ptype_all __read_mostly;    /* Taps */ static struct list_head offload_base __read_mostly; +static struct workqueue_struct *napi_workq __read_mostly; static int netif_rx_internal(struct sk_buff *skb); static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6293

Quoted 7 lines
net/core/dev.c +4 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
    {
        unsigned long flags;
    +    if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+        queue_work(napi_workq, &n->work);
+        return;
+    }
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Quoted 6 lines

Just hoping the thread will eventually run seems optimistic to me.

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

Quoted 6 lines
Quoted 4 lines

We do not want to add code in fast path only for one device. We need something truly generic.

I am not saying only the admin can chose, it is fine if a driver does not give the choice and will simply call netif_threaded_napi_add()

what could make sense if the feature can be disabled / enabled, but it will only affect drivers using the netif_threaded_napi_add call, but it should not affect drivers using the old api in any way since not all drivers will work with this feature.

If we provide something in core NAPI stack, we want to make sure we can test/use it with other drivers.

ethtool, or a /sys/class/net/ethXXX entry could be used.

The argument about not affecting other drivers is misleading, since the patch adds another conditional test in standard NAPI layer.

Lets keep NAPI generic please.

Lets make sure syzbot will find bugs without having to attach a specific mac80211 hardware.

Another concern I have with this patch is that we no longer can contain NIC processing is done on a selected set of cpus (as commanded in /proc/irq/XXX/smp_affinity).
Or can we ?

#8 lore raw

Am 31.07.2020 um 18:36 schrieb Eric Dumazet:

Quoted 122 lines · Sebastian Gottschall
Quoted 120 lines

Am 30.07.2020 um 18:08 schrieb Eric Dumazet:

Quoted 119 lines · Sebastian Gottschall
Quoted 110 lines

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 109 lines · Felix Fietkau
Quoted 96 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

    include/linux/netdevice.h |  23 ++++++
    net/core/dev.c            | 162 ++++++++++++++++++++++++++------------
    2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +22 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
        struct list_head    dev_list;
        struct hlist_node    napi_hash_node;
        unsigned int        napi_id;
+    struct work_struct    work;
    };
      enum {
@@ -357,6 +358,7 @@ enum {
        NAPI_STATE_HASHED,    /* In NAPI hash (busy polling possible) */
        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_THREADED,    /* Use threaded NAPI */
    };
      enum {
@@ -367,6 +369,7 @@ enum {
        NAPIF_STATE_HASHED     = BIT(NAPI_STATE_HASHED),
        NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
        NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+    NAPIF_STATE_THREADED     = BIT(NAPI_STATE_THREADED),
    };
      enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
    void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
                int (*poll)(struct napi_struct *, int), int weight);
    +/**
+ *    netif_threaded_napi_add - initialize a NAPI context
+ *    @dev:  network device
+ *    @napi: NAPI context
+ *    @poll: polling function
+ *    @weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */
+static inline void netif_threaded_napi_add(struct net_device *dev,
+                       struct napi_struct *napi,
+                       int (*poll)(struct napi_struct *, int),
+                       int weight)
+{
+    set_bit(NAPI_STATE_THREADED, &napi->state);
+    netif_napi_add(dev, napi, poll, weight);
+}
+
    /**
     *    netif_tx_napi_add - initialize a NAPI context
     *    @dev:  network device
net/core/dev.c +1 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
    struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
    struct list_head ptype_all __read_mostly;    /* Taps */
    static struct list_head offload_base __read_mostly;
+static struct workqueue_struct *napi_workq __read_mostly;
      static int netif_rx_internal(struct sk_buff *skb);
    static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6293

Quoted 7 lines
net/core/dev.c +4 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
    {
        unsigned long flags;
    +    if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+        queue_work(napi_workq, &n->work);
+        return;
+    }
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Just hoping the thread will eventually run seems optimistic to me.

Quoted 6 lines
Quoted 3 lines

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

We do not want to add code in fast path only for one device. We need something truly generic.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

I am not saying only the admin can chose, it is fine if a driver does not give the choice and will simply call netif_threaded_napi_add()

what could make sense if the feature can be disabled / enabled, but it will only affect drivers using the netif_threaded_napi_add call, but it should not affect drivers using the old api in any way since not all drivers will work with this feature.

If we provide something in core NAPI stack, we want to make sure we can test/use it with other drivers.

ethtool, or a /sys/class/net/ethXXX entry could be used.

but this doesnt work for wifi drivers. since wifi drivers are using dummy netdev devices. we are running in circles here
i mean a sane way could be also that dummy netdev devices are present in sysfs too which is not the case right now.
so changing the api, so the driver is forced to set sane virtual dummy netdev name (like the driver name for instance). so it can be accessed by sysfs.

Quoted 7 lines

The argument about not affecting other drivers is misleading, since the patch adds another conditional test in standard NAPI layer.

Lets keep NAPI generic please.

Lets make sure syzbot will find bugs without having to attach a specific mac80211 hardware.

the patch is not mac80211 specific. i  tested it already with network drivers. it is generic.

Quoted 4 lines

Another concern I have with this patch is that we no longer can contain NIC processing is done on a selected set of cpus (as commanded in /proc/irq/XXX/smp_affinity).
Or can we ?

i had this discussion already with felix in a phonecall last week. kthread vs. workq. his oppinion is that workq works more effective than kthread's since kthreads required application support for good balancing like irqbalance. personally i have no real oppinion here. the good point on kthreads is that i'm able to track the system load per thread with simple process watching. and its possible to force the thread on a specific cpuset.

the good thing on workq is, its more simple to implement and usually more risc free, even if i havent seen any problems with kthreads. maybe felix should say something here about this. the background of this patch is simply performance. especially on embedded devices. it started with a ath10k patch which was introducing napi threading for a specific chipset which leaded to some research on my side until i found a historic napi kthread patch from 2016 which you have denied at that time. i tested that patch and saw heavy performance boost for ath10k. which leaded to this workq patch at the end

see also this discussion here. (you may remember this discussion since you where involved in it)
https://www.mail-archive.com/[email protected]/msg1142611.html

Sebastian

Quoted 2 lines
#9 lore raw
Quoted 136 lines · Sebastian Gottschall

Am 31.07.2020 um 18:36 schrieb Eric Dumazet:

Quoted 122 lines · Sebastian Gottschall
Quoted 120 lines

Am 30.07.2020 um 18:08 schrieb Eric Dumazet:

Quoted 119 lines · Sebastian Gottschall
Quoted 110 lines

Am 29.07.2020 um 19:44 schrieb Eric Dumazet:

Quoted 109 lines · Felix Fietkau
Quoted 96 lines

For some drivers (especially 802.11 drivers), doing a lot of work in the NAPI poll function does not perform well. Since NAPI poll is bound to the CPU it was scheduled from, we can easily end up with a few very busy CPUs spending most of their time in softirq/ksoftirqd and some idle ones.

Introduce threaded NAPI for such drivers based on a workqueue. The API is the same except for using netif_threaded_napi_add instead of netif_napi_add.

In my tests with mt76 on MT7621 using threaded NAPI + a thread for tx scheduling improves LAN->WLAN bridging throughput by 10-50%. Throughput without threaded NAPI is wildly inconsistent, depending on the CPU that runs the tx scheduling thread.

With threaded NAPI, throughput seems stable and consistent (and higher than the best results I got without it).

Based on a patch by Hillf Danton

---
Changes since RFC v2:
- fix unused but set variable reported by kbuild test robot

Changes since RFC:
- disable softirq around threaded poll functions
- reuse most parts of napi_poll()
- fix re-schedule condition

    include/linux/netdevice.h |  23 ++++++
    net/core/dev.c            | 162 ++++++++++++++++++++++++++------------

    2 files changed, 133 insertions(+), 52 deletions(-)
include/linux/netdevice.h +1 -0
index ac2cd3f49aba..3a39211c7598 100644
@@ -347,6 +347,7 @@ struct napi_struct {
         struct list_head    dev_list;
         struct hlist_node    napi_hash_node;
         unsigned int        napi_id;
+    struct work_struct    work;
     };
       enum {
@@ -357,6 +358,7 @@ enum {

        NAPI_STATE_HASHED,    /* In NAPI hash (busy polling possible) */ 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_THREADED,    /* Use threaded NAPI */
    };
      enum {

include/linux/netdevice.h +1 -0
@@ -367,6 +369,7 @@ enum {
         NAPIF_STATE_HASHED     = BIT(NAPI_STATE_HASHED),
         NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
         NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
+    NAPIF_STATE_THREADED     = BIT(NAPI_STATE_THREADED),
     };
       enum gro_result {
@@ -2315,6 +2318,26 @@ static inline void *netdev_priv(const struct net_device *dev)
     void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
                 int (*poll)(struct napi_struct *, int), int weight);

    +/**
+ *    netif_threaded_napi_add - initialize a NAPI context
+ *    @dev:  network device
+ *    @napi: NAPI context
+ *    @poll: polling function
+ *    @weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * with CPU intensive poll functions.
+ * This will schedule polling from a high priority workqueue that
+ */

include/linux/netdevice.h +9 -0
+static inline void netif_threaded_napi_add(struct net_device *dev,
+                       struct napi_struct *napi,
+                       int (*poll)(struct napi_struct *, int),
+                       int weight)
+{
+    set_bit(NAPI_STATE_THREADED, &napi->state);
+    netif_napi_add(dev, napi, poll, weight);
+}
+

    /**
     *    netif_tx_napi_add - initialize a NAPI context *    @dev:  network device

net/core/dev.c +0 -0
index 19f1abc26fcd..11b027f3a2b9 100644
@@ -158,6 +158,7 @@ static DEFINE_SPINLOCK(offload_lock);
     struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;

    struct list_head ptype_all __read_mostly;    /* Taps */ static struct list_head offload_base __read_mostly; +static struct workqueue_struct *napi_workq __read_mostly; static int netif_rx_internal(struct sk_buff *skb); static int call_netdevice_notifiers_info(unsigned long val,

net/core/dev.c void __napi_schedule(struct napi_struct *n) · line 6293

Quoted 7 lines
net/core/dev.c +4 -0
@@ -6286,6 +6287,11 @@ void __napi_schedule(struct napi_struct *n)
     {
         unsigned long flags;
     +    if (test_bit(NAPI_STATE_THREADED, &n->state)) {
+        queue_work(napi_workq, &n->work);
+        return;
+    }
+

Where is the corresponding cancel_work_sync() or flush_work() at device dismantle ?

Just hoping the thread will eventually run seems optimistic to me.

Quoted 6 lines
Quoted 3 lines

Quite frankly, I do believe this STATE_THREADED status should be a generic NAPI attribute that can be changed dynamically, at admin request, instead of having to change/recompile a driver.

thats not that easy. wifi devices do use dummy netdev devices. they are not visible to sysfs and other administrative options. so changing it would just be possible if a special mac80211 based control would be implemented for these drivers. for standard netdev devices it isnt a big thing to implement a administrative control by sysfs (if you are talking about such a feature)

We do not want to add code in fast path only for one device. We need something truly generic.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

I am not saying only the admin can chose, it is fine if a driver does not give the choice and will simply call netif_threaded_napi_add()

what could make sense if the feature can be disabled / enabled, but it will only affect drivers using the netif_threaded_napi_add call, but it should not affect drivers using the old api in any way since not all drivers will work with this feature.

If we provide something in core NAPI stack, we want to make sure we can test/use it with other drivers.

ethtool, or a /sys/class/net/ethXXX entry could be used.

but this doesnt work for wifi drivers. since wifi drivers are using dummy netdev devices. we are running in circles here
i mean a sane way could be also that dummy netdev devices are present in sysfs too which is not the case right now.
so changing the api, so the driver is forced to set sane virtual dummy netdev name (like the driver name for instance). so it can be accessed by sysfs.

Quoted 1 line
Quoted 7 lines
Quoted 6 lines

The argument about not affecting other drivers is misleading, since the patch adds another conditional test in standard NAPI layer.

Lets keep NAPI generic please.

Lets make sure syzbot will find bugs without having to attach a specific mac80211 hardware.

the patch is not mac80211 specific. i  tested it already with network drivers. it is generic.

Not really generic as you have to compile a new driver/kernel, since a driver will either call netif_threaded_napi_add() or netif_napi_add().

Just to be extra clear : We have reasons to like the proposal, and want to use it, without manual changes in NAPI drivers.

Quoted 7 lines
Quoted 4 lines

Another concern I have with this patch is that we no longer can contain NIC processing is done on a selected set of cpus (as commanded in /proc/irq/XXX/smp_affinity).
Or can we ?

i had this discussion already with felix in a phonecall last week. kthread vs. workq. his oppinion is that workq works more effective than kthread's since kthreads required application support for good balancing like irqbalance. personally i have no real oppinion here. the good point on kthreads is that i'm able to track the system load per thread with simple process watching. and its possible to force the thread on a specific cpuset.

In either cases we want to be able tune things from scripts without requesting human looking at a process list :)

Hint : patch does not use WQ_SYSFS