[PATCH net-next v2 0/3] r8169: use new macros from netdev_queues.h

[PATCH net-next v2 0/3] r8169: use new macros from netdev_queues.h

[Heiner Kallweit]

Add one missing subqueue version of the macros, and use the new macros
in r8169 to simplify the code.

Heiner Kallweit (3):
  net: add macro netif_subqueue_completed_wake
  r8169: use new macro netif_subqueue_maybe_stop in rtl8169_start_xmit
  r8169: use new macro netif_subqueue_completed_wake in the tx cleanup path

v2:
- patch 2: ring doorbell if queue was stopped

 drivers/net/ethernet/realtek/r8169_main.c | 51 ++++++-----------------
 include/net/netdev_queues.h               | 10 +++++
 2 files changed, 23 insertions(+), 38 deletions(-)

-- 
2.40.0

[PATCH net-next v2 1/3] net: add macro netif_subqueue_completed_wake

[Heiner Kallweit]

Add netif_subqueue_completed_wake, complementing the subqueue versions
netif_subqueue_try_stop and netif_subqueue_maybe_stop.

Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 include/net/netdev_queues.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/net/netdev_queues.h b/include/net/netdev_queues.h
index b26fdb441..d68b0a483 100644
--- a/include/net/netdev_queues.h
+++ b/include/net/netdev_queues.h
@@ -160,4 +160,14 @@ netdev_txq_completed_mb(struct netdev_queue *dev_queue,
 		netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs); \
 	})
 
+#define netif_subqueue_completed_wake(dev, idx, pkts, bytes,		\
+				      get_desc, start_thrs)		\
+	({								\
+		struct netdev_queue *txq;				\
+									\
+		txq = netdev_get_tx_queue(dev, idx);			\
+		netif_txq_completed_wake(txq, pkts, bytes,		\
+					 get_desc, start_thrs);		\
+	})
+
 #endif
-- 
2.40.0

[PATCH net-next v2 2/3] r8169: use new macro netif_subqueue_maybe_stop in rtl8169_start_xmit

[Heiner Kallweit]

Use new net core macro netif_subqueue_maybe_stop in the start_xmit path
to simplify the code. Whilst at it, set the tx queue start threshold to
twice the stop threshold. Before values were the same, resulting in
stopping/starting the queue more often than needed.

v2:
- ring doorbell if queue was stopped

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 drivers/net/ethernet/realtek/r8169_main.c | 39 +++++++----------------
 1 file changed, 11 insertions(+), 28 deletions(-)

diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index 9f8357bbc..fff44d46b 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -30,6 +30,7 @@
 #include <linux/ipv6.h>
 #include <asm/unaligned.h>
 #include <net/ip6_checksum.h>
+#include <net/netdev_queues.h>
 
 #include "r8169.h"
 #include "r8169_firmware.h"
@@ -68,6 +69,8 @@
 #define NUM_RX_DESC	256	/* Number of Rx descriptor registers */
 #define R8169_TX_RING_BYTES	(NUM_TX_DESC * sizeof(struct TxDesc))
 #define R8169_RX_RING_BYTES	(NUM_RX_DESC * sizeof(struct RxDesc))
+#define R8169_TX_STOP_THRS	(MAX_SKB_FRAGS + 1)
+#define R8169_TX_START_THRS	(2 * R8169_TX_STOP_THRS)
 
 #define OCP_STD_PHY_BASE	0xa400
 
@@ -4162,13 +4165,9 @@ static bool rtl8169_tso_csum_v2(struct rtl8169_private *tp,
 	return true;
 }
 
-static bool rtl_tx_slots_avail(struct rtl8169_private *tp)
+static unsigned int rtl_tx_slots_avail(struct rtl8169_private *tp)
 {
-	unsigned int slots_avail = READ_ONCE(tp->dirty_tx) + NUM_TX_DESC
-					- READ_ONCE(tp->cur_tx);
-
-	/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
-	return slots_avail > MAX_SKB_FRAGS;
+	return READ_ONCE(tp->dirty_tx) + NUM_TX_DESC - READ_ONCE(tp->cur_tx);
 }
 
 /* Versions RTL8102e and from RTL8168c onwards support csum_v2 */
@@ -4245,27 +4244,10 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
 
 	WRITE_ONCE(tp->cur_tx, tp->cur_tx + frags + 1);
 
-	stop_queue = !rtl_tx_slots_avail(tp);
-	if (unlikely(stop_queue)) {
-		/* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
-		 * not miss a ring update when it notices a stopped queue.
-		 */
-		smp_wmb();
-		netif_stop_queue(dev);
-		/* Sync with rtl_tx:
-		 * - publish queue status and cur_tx ring index (write barrier)
-		 * - refresh dirty_tx ring index (read barrier).
-		 * May the current thread have a pessimistic view of the ring
-		 * status and forget to wake up queue, a racing rtl_tx thread
-		 * can't.
-		 */
-		smp_mb__after_atomic();
-		if (rtl_tx_slots_avail(tp))
-			netif_start_queue(dev);
-		door_bell = true;
-	}
-
-	if (door_bell)
+	stop_queue = !netif_subqueue_maybe_stop(dev, 0, rtl_tx_slots_avail(tp),
+						R8169_TX_STOP_THRS,
+						R8169_TX_START_THRS);
+	if (door_bell || stop_queue)
 		rtl8169_doorbell(tp);
 
 	return NETDEV_TX_OK;
@@ -4400,7 +4382,8 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
 		 * ring status.
 		 */
 		smp_store_mb(tp->dirty_tx, dirty_tx);
-		if (netif_queue_stopped(dev) && rtl_tx_slots_avail(tp))
+		if (netif_queue_stopped(dev) &&
+		    rtl_tx_slots_avail(tp) >= R8169_TX_START_THRS)
 			netif_wake_queue(dev);
 		/*
 		 * 8168 hack: TxPoll requests are lost when the Tx packets are
-- 
2.40.0

[PATCH net-next v2 3/3] r8169: use new macro netif_subqueue_completed_wake in the tx cleanup path

[Heiner Kallweit]

Use new net core macro netif_subqueue_completed_wake to simplify
the code of the tx cleanup path.

Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 drivers/net/ethernet/realtek/r8169_main.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)

diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index fff44d46b..a7e376e7e 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -4371,20 +4371,12 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
 	}
 
 	if (tp->dirty_tx != dirty_tx) {
-		netdev_completed_queue(dev, pkts_compl, bytes_compl);
 		dev_sw_netstats_tx_add(dev, pkts_compl, bytes_compl);
+		WRITE_ONCE(tp->dirty_tx, dirty_tx);
 
-		/* Sync with rtl8169_start_xmit:
-		 * - publish dirty_tx ring index (write barrier)
-		 * - refresh cur_tx ring index and queue status (read barrier)
-		 * May the current thread miss the stopped queue condition,
-		 * a racing xmit thread can only have a right view of the
-		 * ring status.
-		 */
-		smp_store_mb(tp->dirty_tx, dirty_tx);
-		if (netif_queue_stopped(dev) &&
-		    rtl_tx_slots_avail(tp) >= R8169_TX_START_THRS)
-			netif_wake_queue(dev);
+		netif_subqueue_completed_wake(dev, 0, pkts_compl, bytes_compl,
+					      rtl_tx_slots_avail(tp),
+					      R8169_TX_START_THRS);
 		/*
 		 * 8168 hack: TxPoll requests are lost when the Tx packets are
 		 * too close. Let's kick an extra TxPoll request when a burst
-- 
2.40.0

Re: [PATCH net-next v2 2/3] r8169: use new macro netif_subqueue_maybe_stop in rtl8169_start_xmit

[Leon Romanovsky]

On Sat, Apr 15, 2023 at 09:22:11AM +0200, Heiner Kallweit wrote:
> Use new net core macro netif_subqueue_maybe_stop in the start_xmit path
> to simplify the code. Whilst at it, set the tx queue start threshold to
> twice the stop threshold. Before values were the same, resulting in
> stopping/starting the queue more often than needed.
> 
> v2:
> - ring doorbell if queue was stopped

Please put changelog under "---" markup, below tags section.

Thanks

> 
> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
> ---
>  drivers/net/ethernet/realtek/r8169_main.c | 39 +++++++----------------
>  1 file changed, 11 insertions(+), 28 deletions(-)
> 
> diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
> index 9f8357bbc..fff44d46b 100644
> --- a/drivers/net/ethernet/realtek/r8169_main.c
> +++ b/drivers/net/ethernet/realtek/r8169_main.c
> @@ -30,6 +30,7 @@
>  #include <linux/ipv6.h>
>  #include <asm/unaligned.h>
>  #include <net/ip6_checksum.h>
> +#include <net/netdev_queues.h>
>  
>  #include "r8169.h"
>  #include "r8169_firmware.h"
> @@ -68,6 +69,8 @@
>  #define NUM_RX_DESC	256	/* Number of Rx descriptor registers */
>  #define R8169_TX_RING_BYTES	(NUM_TX_DESC * sizeof(struct TxDesc))
>  #define R8169_RX_RING_BYTES	(NUM_RX_DESC * sizeof(struct RxDesc))
> +#define R8169_TX_STOP_THRS	(MAX_SKB_FRAGS + 1)
> +#define R8169_TX_START_THRS	(2 * R8169_TX_STOP_THRS)
>  
>  #define OCP_STD_PHY_BASE	0xa400
>  
> @@ -4162,13 +4165,9 @@ static bool rtl8169_tso_csum_v2(struct rtl8169_private *tp,
>  	return true;
>  }
>  
> -static bool rtl_tx_slots_avail(struct rtl8169_private *tp)
> +static unsigned int rtl_tx_slots_avail(struct rtl8169_private *tp)
>  {
> -	unsigned int slots_avail = READ_ONCE(tp->dirty_tx) + NUM_TX_DESC
> -					- READ_ONCE(tp->cur_tx);
> -
> -	/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
> -	return slots_avail > MAX_SKB_FRAGS;
> +	return READ_ONCE(tp->dirty_tx) + NUM_TX_DESC - READ_ONCE(tp->cur_tx);
>  }
>  
>  /* Versions RTL8102e and from RTL8168c onwards support csum_v2 */
> @@ -4245,27 +4244,10 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
>  
>  	WRITE_ONCE(tp->cur_tx, tp->cur_tx + frags + 1);
>  
> -	stop_queue = !rtl_tx_slots_avail(tp);
> -	if (unlikely(stop_queue)) {
> -		/* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
> -		 * not miss a ring update when it notices a stopped queue.
> -		 */
> -		smp_wmb();
> -		netif_stop_queue(dev);
> -		/* Sync with rtl_tx:
> -		 * - publish queue status and cur_tx ring index (write barrier)
> -		 * - refresh dirty_tx ring index (read barrier).
> -		 * May the current thread have a pessimistic view of the ring
> -		 * status and forget to wake up queue, a racing rtl_tx thread
> -		 * can't.
> -		 */
> -		smp_mb__after_atomic();
> -		if (rtl_tx_slots_avail(tp))
> -			netif_start_queue(dev);
> -		door_bell = true;
> -	}
> -
> -	if (door_bell)
> +	stop_queue = !netif_subqueue_maybe_stop(dev, 0, rtl_tx_slots_avail(tp),
> +						R8169_TX_STOP_THRS,
> +						R8169_TX_START_THRS);
> +	if (door_bell || stop_queue)
>  		rtl8169_doorbell(tp);
>  
>  	return NETDEV_TX_OK;
> @@ -4400,7 +4382,8 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
>  		 * ring status.
>  		 */
>  		smp_store_mb(tp->dirty_tx, dirty_tx);
> -		if (netif_queue_stopped(dev) && rtl_tx_slots_avail(tp))
> +		if (netif_queue_stopped(dev) &&
> +		    rtl_tx_slots_avail(tp) >= R8169_TX_START_THRS)
>  			netif_wake_queue(dev);
>  		/*
>  		 * 8168 hack: TxPoll requests are lost when the Tx packets are
> -- 
> 2.40.0
> 
> 

Re: [PATCH net-next v2 2/3] r8169: use new macro netif_subqueue_maybe_stop in rtl8169_start_xmit

[Heiner Kallweit]

On 16.04.2023 12:20, Leon Romanovsky wrote:
> On Sat, Apr 15, 2023 at 09:22:11AM +0200, Heiner Kallweit wrote:
>> Use new net core macro netif_subqueue_maybe_stop in the start_xmit path
>> to simplify the code. Whilst at it, set the tx queue start threshold to
>> twice the stop threshold. Before values were the same, resulting in
>> stopping/starting the queue more often than needed.
>>
>> v2:
>> - ring doorbell if queue was stopped
> 
> Please put changelog under "---" markup, below tags section.
> 
I know that this would be the standard. IIRC Dave once requested to
make the change log part of the commit message.

> Thanks
> 

Re: [PATCH net-next v2 2/3] r8169: use new macro netif_subqueue_maybe_stop in rtl8169_start_xmit

[Leon Romanovsky]

On Sun, Apr 16, 2023 at 01:33:11PM +0200, Heiner Kallweit wrote:
> On 16.04.2023 12:20, Leon Romanovsky wrote:
> > On Sat, Apr 15, 2023 at 09:22:11AM +0200, Heiner Kallweit wrote:
> >> Use new net core macro netif_subqueue_maybe_stop in the start_xmit path
> >> to simplify the code. Whilst at it, set the tx queue start threshold to
> >> twice the stop threshold. Before values were the same, resulting in
> >> stopping/starting the queue more often than needed.
> >>
> >> v2:
> >> - ring doorbell if queue was stopped
> > 
> > Please put changelog under "---" markup, below tags section.
> > 
> I know that this would be the standard. IIRC Dave once requested to
> make the change log part of the commit message.

I can imagine how it was useful before lore.kernel.org appeared.

Thanks

> 
> > Thanks
> > 
>