Re: [PATCH net v2] lan743x: fix rx_napi_poll/interrupt ping-pong
Am 11.12.2020 um 13:43 schrieb Sven Van Asbroeck: > Hi Heiner, > > On Thu, Dec 10, 2020 at 2:32 AM Heiner Kallweit wrote: >> >> >> In addition you could play with sysfs attributes >> /sys/class/net//gro_flush_timeout >> /sys/class/net//napi_defer_hard_irqs > > Interesting, I will look into that. > I run a 1Gbit chip with gro_flush_timeout = 2 and napi_defer_hard_irqs = 1. This helped to reduce interrupt load significantly under iperf3 (w/o interrupt coalescing at chip level) >>> @@ -2407,7 +2409,7 @@ static int lan743x_rx_open(struct lan743x_rx *rx) >>> >>> netif_napi_add(adapter->netdev, >>> >napi, lan743x_rx_napi_poll, >>> -rx->ring_size - 1); >>> +64); >> >> This value isn't completely arbitrary. >> Better use constant NAPI_POLL_WEIGHT. >> > > Thank you, I will change it in the next patch version. >
Re: [PATCH net v2] lan743x: fix rx_napi_poll/interrupt ping-pong
Hi Heiner, On Thu, Dec 10, 2020 at 2:32 AM Heiner Kallweit wrote: > > > In addition you could play with sysfs attributes > /sys/class/net//gro_flush_timeout > /sys/class/net//napi_defer_hard_irqs Interesting, I will look into that. > > @@ -2407,7 +2409,7 @@ static int lan743x_rx_open(struct lan743x_rx *rx) > > > > netif_napi_add(adapter->netdev, > > >napi, lan743x_rx_napi_poll, > > -rx->ring_size - 1); > > +64); > > This value isn't completely arbitrary. > Better use constant NAPI_POLL_WEIGHT. > Thank you, I will change it in the next patch version.
Re: [PATCH net v2] lan743x: fix rx_napi_poll/interrupt ping-pong
Am 10.12.2020 um 04:55 schrieb Sven Van Asbroeck:
> From: Sven Van Asbroeck
>
> Even if there is more rx data waiting on the chip, the rx napi poll fn
> will never run more than once - it will always read a few buffers, then
> bail out and re-arm interrupts. Which results in ping-pong between napi
> and interrupt.
>
> This defeats the purpose of napi, and is bad for performance.
>
> Fix by making the rx napi poll behave identically to other ethernet
> drivers:
> 1. initialize rx napi polling with an arbitrary budget (64).
> 2. in the polling fn, return full weight if rx queue is not depleted,
>this tells the napi core to "keep polling".
> 3. update the rx tail ("ring the doorbell") once for every 8 processed
>rx ring buffers.
>
> Thanks to Jakub Kicinski, Eric Dumazet and Andrew Lunn for their expert
> opinions and suggestions.
>
> Tested with 20 seconds of full bandwidth receive (iperf3):
> rx irqs softirqs(NET_RX)
> -
> before 2382733620
> after 129 4081
>
In addition you could play with sysfs attributes
/sys/class/net//gro_flush_timeout
/sys/class/net//napi_defer_hard_irqs
> Tested-by: Sven Van Asbroeck # lan7430
> Fixes: 23f0703c125be ("lan743x: Add main source files for new lan743x driver")
> Signed-off-by: Sven Van Asbroeck
> ---
>
> Tree: git://git.kernel.org/pub/scm/linux/kernel/git/davem/net.git #
> b7e4ba9a91df
>
> To: Bryan Whitehead
> To: Microchip Linux Driver Support
> To: "David S. Miller"
> To: Jakub Kicinski
> Cc: Andrew Lunn
> Cc: net...@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
>
> drivers/net/ethernet/microchip/lan743x_main.c | 44 ++-
> 1 file changed, 23 insertions(+), 21 deletions(-)
>
> diff --git a/drivers/net/ethernet/microchip/lan743x_main.c
> b/drivers/net/ethernet/microchip/lan743x_main.c
> index 87b6c59a1e03..30ec308b9a4c 100644
> --- a/drivers/net/ethernet/microchip/lan743x_main.c
> +++ b/drivers/net/ethernet/microchip/lan743x_main.c
> @@ -1964,6 +1964,14 @@ static struct sk_buff *lan743x_rx_allocate_skb(struct
> lan743x_rx *rx)
> length, GFP_ATOMIC | GFP_DMA);
> }
>
> +static void lan743x_rx_update_tail(struct lan743x_rx *rx, int index)
> +{
> + /* update the tail once per 8 descriptors */
> + if ((index & 7) == 7)
> + lan743x_csr_write(rx->adapter, RX_TAIL(rx->channel_number),
> + index);
> +}
> +
> static int lan743x_rx_init_ring_element(struct lan743x_rx *rx, int index,
> struct sk_buff *skb)
> {
> @@ -1994,6 +2002,7 @@ static int lan743x_rx_init_ring_element(struct
> lan743x_rx *rx, int index,
> descriptor->data0 = (RX_DESC_DATA0_OWN_ |
> (length & RX_DESC_DATA0_BUF_LENGTH_MASK_));
> skb_reserve(buffer_info->skb, RX_HEAD_PADDING);
> + lan743x_rx_update_tail(rx, index);
>
> return 0;
> }
> @@ -2012,6 +2021,7 @@ static void lan743x_rx_reuse_ring_element(struct
> lan743x_rx *rx, int index)
> descriptor->data0 = (RX_DESC_DATA0_OWN_ |
> ((buffer_info->buffer_length) &
> RX_DESC_DATA0_BUF_LENGTH_MASK_));
> + lan743x_rx_update_tail(rx, index);
> }
>
> static void lan743x_rx_release_ring_element(struct lan743x_rx *rx, int index)
> @@ -2223,34 +2233,26 @@ static int lan743x_rx_napi_poll(struct napi_struct
> *napi, int weight)
> struct lan743x_rx *rx = container_of(napi, struct lan743x_rx, napi);
> struct lan743x_adapter *adapter = rx->adapter;
> u32 rx_tail_flags = 0;
> - int count;
> + int count, result;
>
> if (rx->vector_flags & LAN743X_VECTOR_FLAG_SOURCE_STATUS_W2C) {
> /* clear int status bit before reading packet */
> lan743x_csr_write(adapter, DMAC_INT_STS,
> DMAC_INT_BIT_RXFRM_(rx->channel_number));
> }
> - count = 0;
> - while (count < weight) {
> - int rx_process_result = lan743x_rx_process_packet(rx);
> -
> - if (rx_process_result == RX_PROCESS_RESULT_PACKET_RECEIVED) {
> - count++;
> - } else if (rx_process_result ==
> - RX_PROCESS_RESULT_NOTHING_TO_DO) {
> + for (count = 0; count < weight; count++) {
> + result = lan743x_rx_process_packet(rx);
> + if (result == RX_PROCESS_RESULT_NOTHING_TO_DO)
> break;
> - } else if (rx_process_result ==
> - RX_PROCESS_RESULT_PACKET_DROPPED) {
> - continue;
> - }
> }
> rx->frame_count += count;
> - if (count == weight)
> - goto done;
> + if (count == weight || result == RX_PROCESS_RESULT_PACKET_RECEIVED)
> + return weight;
>
> if (!napi_complete_done(napi, count))
> - goto done;
> + return count;
>
>
[PATCH net v2] lan743x: fix rx_napi_poll/interrupt ping-pong
From: Sven Van Asbroeck
Even if there is more rx data waiting on the chip, the rx napi poll fn
will never run more than once - it will always read a few buffers, then
bail out and re-arm interrupts. Which results in ping-pong between napi
and interrupt.
This defeats the purpose of napi, and is bad for performance.
Fix by making the rx napi poll behave identically to other ethernet
drivers:
1. initialize rx napi polling with an arbitrary budget (64).
2. in the polling fn, return full weight if rx queue is not depleted,
this tells the napi core to "keep polling".
3. update the rx tail ("ring the doorbell") once for every 8 processed
rx ring buffers.
Thanks to Jakub Kicinski, Eric Dumazet and Andrew Lunn for their expert
opinions and suggestions.
Tested with 20 seconds of full bandwidth receive (iperf3):
rx irqs softirqs(NET_RX)
-
before 2382733620
after 129 4081
Tested-by: Sven Van Asbroeck # lan7430
Fixes: 23f0703c125be ("lan743x: Add main source files for new lan743x driver")
Signed-off-by: Sven Van Asbroeck
---
Tree: git://git.kernel.org/pub/scm/linux/kernel/git/davem/net.git # b7e4ba9a91df
To: Bryan Whitehead
To: Microchip Linux Driver Support
To: "David S. Miller"
To: Jakub Kicinski
Cc: Andrew Lunn
Cc: net...@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
drivers/net/ethernet/microchip/lan743x_main.c | 44 ++-
1 file changed, 23 insertions(+), 21 deletions(-)
diff --git a/drivers/net/ethernet/microchip/lan743x_main.c
b/drivers/net/ethernet/microchip/lan743x_main.c
index 87b6c59a1e03..30ec308b9a4c 100644
--- a/drivers/net/ethernet/microchip/lan743x_main.c
+++ b/drivers/net/ethernet/microchip/lan743x_main.c
@@ -1964,6 +1964,14 @@ static struct sk_buff *lan743x_rx_allocate_skb(struct
lan743x_rx *rx)
length, GFP_ATOMIC | GFP_DMA);
}
+static void lan743x_rx_update_tail(struct lan743x_rx *rx, int index)
+{
+ /* update the tail once per 8 descriptors */
+ if ((index & 7) == 7)
+ lan743x_csr_write(rx->adapter, RX_TAIL(rx->channel_number),
+ index);
+}
+
static int lan743x_rx_init_ring_element(struct lan743x_rx *rx, int index,
struct sk_buff *skb)
{
@@ -1994,6 +2002,7 @@ static int lan743x_rx_init_ring_element(struct lan743x_rx
*rx, int index,
descriptor->data0 = (RX_DESC_DATA0_OWN_ |
(length & RX_DESC_DATA0_BUF_LENGTH_MASK_));
skb_reserve(buffer_info->skb, RX_HEAD_PADDING);
+ lan743x_rx_update_tail(rx, index);
return 0;
}
@@ -2012,6 +2021,7 @@ static void lan743x_rx_reuse_ring_element(struct
lan743x_rx *rx, int index)
descriptor->data0 = (RX_DESC_DATA0_OWN_ |
((buffer_info->buffer_length) &
RX_DESC_DATA0_BUF_LENGTH_MASK_));
+ lan743x_rx_update_tail(rx, index);
}
static void lan743x_rx_release_ring_element(struct lan743x_rx *rx, int index)
@@ -2223,34 +2233,26 @@ static int lan743x_rx_napi_poll(struct napi_struct
*napi, int weight)
struct lan743x_rx *rx = container_of(napi, struct lan743x_rx, napi);
struct lan743x_adapter *adapter = rx->adapter;
u32 rx_tail_flags = 0;
- int count;
+ int count, result;
if (rx->vector_flags & LAN743X_VECTOR_FLAG_SOURCE_STATUS_W2C) {
/* clear int status bit before reading packet */
lan743x_csr_write(adapter, DMAC_INT_STS,
DMAC_INT_BIT_RXFRM_(rx->channel_number));
}
- count = 0;
- while (count < weight) {
- int rx_process_result = lan743x_rx_process_packet(rx);
-
- if (rx_process_result == RX_PROCESS_RESULT_PACKET_RECEIVED) {
- count++;
- } else if (rx_process_result ==
- RX_PROCESS_RESULT_NOTHING_TO_DO) {
+ for (count = 0; count < weight; count++) {
+ result = lan743x_rx_process_packet(rx);
+ if (result == RX_PROCESS_RESULT_NOTHING_TO_DO)
break;
- } else if (rx_process_result ==
- RX_PROCESS_RESULT_PACKET_DROPPED) {
- continue;
- }
}
rx->frame_count += count;
- if (count == weight)
- goto done;
+ if (count == weight || result == RX_PROCESS_RESULT_PACKET_RECEIVED)
+ return weight;
if (!napi_complete_done(napi, count))
- goto done;
+ return count;
+ /* re-arm interrupts, must write to rx tail on some chip variants */
if (rx->vector_flags & LAN743X_VECTOR_FLAG_VECTOR_ENABLE_AUTO_SET)
rx_tail_flags |= RX_TAIL_SET_TOP_INT_VEC_EN_;
if (rx->vector_flags & LAN743X_VECTOR_FLAG_SOURCE_ENABLE_AUTO_SET) {
@@ -2260,10 +2262,10 @@ static
