Skip to contentSkip to conversation

r8169: disable default rx interrupt coalescing on RTL8168

PATCH In review Heiner Kallweit 3 participants 4 messages Mar 30, 2019 – Apr 1, 2019

Subsystems (3)
lore
Share & tools
lore

Heiner Kallweit

author maintainer
#0 lore raw

PATCH 1/1 r8169: disable default rx interrupt coalescing on RTL8168

It was reported that re-introducing ASPM, in combination with RX interrupt coalescing, results in significantly increased packet latency, see [0]. Disabling ASPM or RX interrupt coalescing fixes the issue. Therefore change the driver's default to disable RX interrupt coalescing. Users still have the option to enable RX coalescing via ethtool.

[0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=925496

Fixes a99790bf5c7f ("r8169: Reinstate ASPM Support")
---
 drivers/net/ethernet/realtek/r8169.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/realtek/r8169.c +1 -1
index c9ee1c8eb..8a8a4f70d 100644
@@ -5460,7 +5460,7 @@ static void rtl_hw_start_8168(struct rtl8169_private *tp)
 	tp->cp_cmd |= PktCntrDisable | INTT_1;
 	RTL_W16(tp, CPlusCmd, tp->cp_cmd);
 
-	RTL_W16(tp, IntrMitigate, 0x5151);
+	RTL_W16(tp, IntrMitigate, 0x5100);
 
 	/* Work around for RxFIFO overflow. */
 	if (tp->mac_version == RTL_GIGA_MAC_VER_11) {
Signature · 1 line
2.21.0
maintainer
#1 lore raw

From: Heiner Kallweit <[email protected]>
Date: Sat, 30 Mar 2019 17:13:24 +0100

Quoted 7 lines

It was reported that re-introducing ASPM, in combination with RX interrupt coalescing, results in significantly increased packet latency, see [0]. Disabling ASPM or RX interrupt coalescing fixes the issue. Therefore change the driver's default to disable RX interrupt coalescing. Users still have the option to enable RX coalescing via ethtool.

Quoted 5 lines

[0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=925496

Fixes a99790bf5c7f ("r8169: Reinstate ASPM Support")

Applied and queued up for -stable.

Frankly, I think we were better off with ASPM disabled. But I know a lot of people find that unacceptable.

This chip has some many quirks/warts when ASPM is enabled, look how long it took us to A) get ASPM enabled reasonably and B) discover this bug.

author maintainer
#2 lore raw
Quoted 12 lines · David Miller

From: Heiner Kallweit <[email protected]>
Date: Sat, 30 Mar 2019 17:13:24 +0100

Quoted 9 lines

It was reported that re-introducing ASPM, in combination with RX interrupt coalescing, results in significantly increased packet latency, see [0]. Disabling ASPM or RX interrupt coalescing fixes the issue. Therefore change the driver's default to disable RX interrupt coalescing. Users still have the option to enable RX coalescing via ethtool.

[0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=925496

Quoted 9 lines
Quoted 3 lines
Fixes a99790bf5c7f ("r8169: Reinstate ASPM Support")

Applied and queued up for -stable.

Frankly, I think we were better off with ASPM disabled. But I know a lot of people find that unacceptable.

Yes, I can hear the notebook owners scream already, and complain about reduced battery runtime. It would help to have an opt-in option for ASPM. But a module parameter for it is discouraged, and I see no other acceptable way to control ASPM on the chip side. I could misuse some ethtool callback or add a new tunable. But controlling a PCI feature through ethtool doesn't sound too nice. Maybe I could add a sysfs attribute.

Quoted 4 lines

This chip has some many quirks/warts when ASPM is enabled, look how long it took us to A) get ASPM enabled reasonably and B) discover this bug.

Tested-by on v1 1/1
#3 lore raw
In reply to Heiner Kallweit
Quoted 31 lines · Heiner Kallweit

It was reported that re-introducing ASPM, in combination with RX interrupt coalescing, results in significantly increased packet latency, see [0]. Disabling ASPM or RX interrupt coalescing fixes the issue. Therefore change the driver's default to disable RX interrupt coalescing. Users still have the option to enable RX coalescing via ethtool.

[0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=925496

Fixes a99790bf5c7f ("r8169: Reinstate ASPM Support")
---
 drivers/net/ethernet/realtek/r8169.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/realtek/r8169.c +1 -1
index c9ee1c8eb..8a8a4f70d 100644
@@ -5460,7 +5460,7 @@ static void rtl_hw_start_8168(struct rtl8169_private *tp)
 	tp->cp_cmd |= PktCntrDisable | INTT_1;
 	RTL_W16(tp, CPlusCmd, tp->cp_cmd);
 
-	RTL_W16(tp, IntrMitigate, 0x5151);
+	RTL_W16(tp, IntrMitigate, 0x5100);
 
 	/* Work around for RxFIFO overflow. */
 	if (tp->mac_version == RTL_GIGA_MAC_VER_11) {
Signature · 1 line
2.21.0

I configured my ASUS BIOS to reenable "EPA Power Saving Mode" and tried running the current state of linux-5.0.y with the above patch applied on top.

As expected, packet latency was good immediately after booting. Running "ethtool -c" showed rx-usecs=0, rx-frames=1. Also as expected, manually increasing these to rx-usecs=200, rx-frames=4 caused the latency problem to return.

Thanks for finding a way to address this problem without causing both users of ageing motherboards and modern notebooks to scream.

Mike.