Skip to contentSkip to conversation

net: add and use netdev_sw_irq_coalesce_default_on()

PATCH Applied Heiner Kallweit 2 participants 4 messages Nov 30, 2022 – Dec 5, 2022

Subsystems (3)
lore
Share & tools
lore

Heiner Kallweit

author maintainer
#0 lore raw

cover net: add and use netdev_sw_irq_coalesce_default_on()

There are reports about r8169 not reaching full line speed on certain systems (e.g. SBC's) with a 2.5Gbps link.
There was a time when hardware interrupt coalescing was enabled per default, but this was changed due to ASPM-related issues on few systems.

Meanwhile we have sysfs attributes for controlling kind of "software interrupt coalescing" on the GRO level. However most distros and users don't know about it. So lets set a conservative default for both involved parameters. Users can still override the defaults via sysfs. Don't enable these settings on the fast ethernet chip versions, they are slow enough.

Even with these conservative setting interrupt load on my 1Gbps test system reduced significantly.

Follow Jakub's suggestion and put this functionality into net core so that other MAC drivers can reuse it.

Heiner Kallweit (2):
net: add netdev_sw_irq_coalesce_default_on()
r8169: enable GRO software interrupt coalescing per default

 drivers/net/ethernet/realtek/r8169_main.c |  2 ++
 include/linux/netdevice.h                 |  1 +
 net/core/dev.c                            | 16 ++++++++++++++++
 3 files changed, 19 insertions(+)
Signature · 1 line
2.38.1
author maintainer
#1 lore raw

PATCH 1/2 net: add netdev_sw_irq_coalesce_default_on()

Add a helper for drivers wanting to set SW IRQ coalescing by default. The related sysfs attributes can be used to override the default values.

Follow Jakub's suggestion and put this functionality into net core so that drivers wanting to use software interrupt coalescing per default don't have to open-code it.

Note that this function needs to be called before the netdevice is registered.

---
 include/linux/netdevice.h |  1 +
 net/core/dev.c            | 16 ++++++++++++++++
 2 files changed, 17 insertions(+)
include/linux/netdevice.h +1 -0
index 02a2318da..5be4b6a3b 100644
@@ -78,6 +78,7 @@ struct xdp_buff;
 void synchronize_net(void);
 void netdev_set_default_ethtool_ops(struct net_device *dev,
 				    const struct ethtool_ops *ops);
+void netdev_sw_irq_coalesce_default_on(struct net_device *dev);
 
 /* Backlog congestion levels */
 #define NET_RX_SUCCESS		0	/* keep 'em coming, baby */
net/core/dev.c +16 -0
index 117e830ca..cc6bbc0a3 100644
@@ -10537,6 +10537,22 @@ void netdev_set_default_ethtool_ops(struct net_device *dev,
 }
 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
 
+/**
+ * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
+ * @dev: netdev to enable the IRQ coalescing on
+ *
+ * Sets a conservative default for SW IRQ coalescing. Users can use
+ * sysfs attributes to override the default values.
+ */
+void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
+{
+	WARN_ON(dev->reg_state == NETREG_REGISTERED);
+
+	dev->gro_flush_timeout = 20000;
+	dev->napi_defer_hard_irqs = 1;
+}
+EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
+
 void netdev_freemem(struct net_device *dev)
 {
 	char *addr = (char *)dev - dev->padded;
Signature · 1 line
2.38.1
author maintainer
#2 lore raw

PATCH 2/2 r8169: enable GRO software interrupt coalescing per default

In reply to Heiner Kallweit

There are reports about r8169 not reaching full line speed on certain systems (e.g. SBC's) with a 2.5Gbps link.
There was a time when hardware interrupt coalescing was enabled per default, but this was changed due to ASPM-related issues on few systems. So let's use software interrupt coalescing instead and enable it using new function netdev_sw_irq_coalesce_default_on().

Even with these conservative settings interrupt load on my 1Gbps test system reduced significantly.

---
 drivers/net/ethernet/realtek/r8169_main.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/ethernet/realtek/r8169_main.c +2 -0
index 5bc1181f8..f4e160888 100644
@@ -5282,6 +5282,8 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	dev->hw_features |= NETIF_F_RXALL;
 	dev->hw_features |= NETIF_F_RXFCS;
 
+	netdev_sw_irq_coalesce_default_on(dev);
+
 	/* configure chip for default features */
 	rtl8169_set_features(dev, dev->features);
 
Signature · 1 line
2.38.1
#3 lore raw
In reply to Heiner Kallweit
This series was applied to netdev/net-next.git (master) by David S. Miller <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (master) by David S. Miller <[email protected]>:

Quoted 5 lines · On Wed, 30 Nov 2022 23:25:23 +0100 you wrote:

There are reports about r8169 not reaching full line speed on certain systems (e.g. SBC's) with a 2.5Gbps link.
There was a time when hardware interrupt coalescing was enabled per default, but this was changed due to ASPM-related issues on few systems.

Quoted 8 lines

Meanwhile we have sysfs attributes for controlling kind of "software interrupt coalescing" on the GRO level. However most distros and users don't know about it. So lets set a conservative default for both involved parameters. Users can still override the defaults via sysfs. Don't enable these settings on the fast ethernet chip versions, they are slow enough.

[...]

Here is the summary with links:
- [net-next,1/2] net: add netdev_sw_irq_coalesce_default_on()

    https://git.kernel.org/netdev/net-next/c/d93607082e98
  - [net-next,2/2] r8169: enable GRO software interrupt coalescing per default
    https://git.kernel.org/netdev/net-next/c/42f66a44d837

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html