Skip to contentSkip to conversation

net: phy: marvell: add link status check before enabling phy loopback

PATCH In review Aminuddin Jamaluddin 2 participants 7 messages Aug 25, 2022 – Oct 27, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: phy: marvell: add link status check before enabling phy loopback

Add link status checking in m88e1510_loopback() for 1Gbps link speed and delay for 100ms after phy loopback bit is set before send packet. This is needed to ensure the stability and consistency when running the phy loopback test.

Fixes 020a45aff119 ("net: phy: marvell: add Marvell specific PHY loopback")
Cc <[email protected]> # 5.15.x
---
 drivers/net/phy/marvell.c | 22 ++++++++++++++++------
 include/linux/phy.h       |  3 +++
 2 files changed, 19 insertions(+), 6 deletions(-)
drivers/net/phy/marvell.c +16 -6
index a714150f5e8c..17403acf780d 100644
@@ -1992,6 +1992,7 @@ static int m88e1510_loopback(struct phy_device *phydev, bool enable)
 
 	if (enable) {
 		u16 bmcr_ctl, mscr2_ctl = 0;
+		int val = 0;
 
 		bmcr_ctl = mii_bmcr_encode_fixed(phydev->speed, phydev->duplex);
 
@@ -2015,14 +2016,23 @@ static int m88e1510_loopback(struct phy_device *phydev, bool enable)
 		if (err < 0)
 			return err;
 
-		/* FIXME: Based on trial and error test, it seem 1G need to have
-		 * delay between soft reset and loopback enablement.
-		 */
-		if (phydev->speed == SPEED_1000)
-			msleep(1000);
+		if (phydev->speed == SPEED_1000) {
+			err = phy_read_poll_timeout(phydev, MII_BMSR, val, val & BMSR_LSTATUS,
+						    PHY_LOOP_BACK_SLEEP,
+						    PHY_LOOP_BACK_TIMEOUT, true);
+			if (err)
+				return err;
+		}
 
-		return phy_modify(phydev, MII_BMCR, BMCR_LOOPBACK,
+		err =  phy_modify(phydev, MII_BMCR, BMCR_LOOPBACK,
 				  BMCR_LOOPBACK);
+		if (!err) {
+			/* It takes some time for PHY device to switch
+			 * into/out-of loopback mode.
+			 */
+			msleep(100);
+		}
+		return err;
 	} else {
 		err = phy_modify(phydev, MII_BMCR, BMCR_LOOPBACK, 0);
 		if (err < 0)
include/linux/phy.h +3 -0
index 87638c55d844..b4da968da8e6 100644
@@ -276,6 +276,9 @@ static inline const char *phy_modes(phy_interface_t interface)
 #define PHY_INIT_TIMEOUT	100000
 #define PHY_FORCE_TIMEOUT	10
 
+#define PHY_LOOP_BACK_SLEEP	1000000
+#define PHY_LOOP_BACK_TIMEOUT	8000000
+
 #define PHY_MAX_ADDR	32
 
 /* Used when trying to connect to a specific phy (mii bus id:phy device id) */
Signature · 1 line
2.17.1
maintainer
#1 lore raw

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable) · line 2022

Quoted 13 lines
drivers/net/phy/marvell.c +4 -5
@@ -2015,14 +2016,23 @@ static int m88e1510_loopback(struct phy_device *phydev, bool enable)
 		if (err < 0)
 			return err;
 
-		/* FIXME: Based on trial and error test, it seem 1G need to have
-		 * delay between soft reset and loopback enablement.
-		 */
-		if (phydev->speed == SPEED_1000)
-			msleep(1000);
+		if (phydev->speed == SPEED_1000) {
+			err = phy_read_poll_timeout(phydev, MII_BMSR, val, val & BMSR_LSTATUS,
+						    PHY_LOOP_BACK_SLEEP,
+						    PHY_LOOP_BACK_TIMEOUT, true);

Is this link with itself?

Have you tested this with the cable unplugged?

Review static int m88e1510_loopback(struct phy_device *phydev, bool enable) · line 2024

Quoted 2 lines
(quoted diff) +2 -0
+			if (err)
+				return err;

I'm just trying to ensure we don't end up here with -ETIMEDOUT.

include/linux/phy.h static inline const char *phy_modes(phy_interface_t interface) · line 280

Quoted 3 lines
include/linux/phy.h +2 -0
+#define PHY_LOOP_BACK_SLEEP	1000000
+#define PHY_LOOP_BACK_TIMEOUT	8000000

The kernel seems to be pretty consistent in having loopback as one word.

Andrew

#2 lore raw
Quoted 31 lines

-----Original Message-----
From: Andrew Lunn <[email protected]>
Sent: Thursday, 25 August, 2022 9:27 PM
To: Jamaluddin, Aminuddin <[email protected]>
Cc: Heiner Kallweit <[email protected]>; Russell King <[email protected]>; David S . Miller <[email protected]>; Eric Dumazet <[email protected]>; Jakub Kicinski <[email protected]>; Paolo Abeni <[email protected]>; Ismail, Mohammad Athari <[email protected]>; [email protected]; linux- [email protected]; [email protected]; Tan, Tee Min <[email protected]>; Zulkifli, Muhammad Husaini <[email protected]>
Subject: Re: [PATCH net 1/1] net: phy: marvell: add link status check before enabling phy loopback

Review static int m88e1510_loopback(struct

Quoted 1 line
(quoted diff) +0 -0
@@ -2015,14 +2016,23 @@ static int m88e1510_loopback(struct

phy_device *phydev, bool enable)

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 4 lines
drivers/net/phy/marvell.c +0 -1
 		if (err < 0)
 			return err;
 
-		/* FIXME: Based on trial and error test, it seem 1G need to

have

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 6 lines
drivers/net/phy/marvell.c +2 -4
-		 * delay between soft reset and loopback enablement.
-		 */
-		if (phydev->speed == SPEED_1000)
-			msleep(1000);
+		if (phydev->speed == SPEED_1000) {
+			err = phy_read_poll_timeout(phydev, MII_BMSR,

val, val & BMSR_LSTATUS,

Quoted 5 lines

Review static int m88e1510_loopback(struct · line 2022

Quoted 2 lines
(quoted diff) +2 -0
+						    PHY_LOOP_BACK_SLEEP,
+

PHY_LOOP_BACK_TIMEOUT, true);

Is this link with itself?

Its required cabled plug in, back to back connection.

Quoted 2 lines

Have you tested this with the cable unplugged?

Yes we have and its expected to have the timeout. But the self-test required the link to be up first before it can be run.

Quoted 11 lines

Review static int m88e1510_loopback(struct · line 2024

Quoted 3 lines
(quoted diff) +2 -0
+			if (err)
+				return err;

I'm just trying to ensure we don't end up here with -ETIMEDOUT.

include/linux/phy.h static inline const char *phy_modes(phy_interface_t interface) · line 280

Quoted 3 lines
include/linux/phy.h +2 -0
+#define PHY_LOOP_BACK_SLEEP	1000000
+#define PHY_LOOP_BACK_TIMEOUT	8000000

The kernel seems to be pretty consistent in having loopback as one word.

Noted will update in v2.

Quoted 2 lines

Andrew

maintainer
#3 lore raw
Quoted 14 lines
Quoted 14 lines

Review static int m88e1510_loopback(struct

Quoted 1 line
(quoted diff) +0 -0
@@ -2015,14 +2016,23 @@ static int m88e1510_loopback(struct

phy_device *phydev, bool enable)

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 4 lines
drivers/net/phy/marvell.c +0 -1
 		if (err < 0)
 			return err;
 
-		/* FIXME: Based on trial and error test, it seem 1G need to

have

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 6 lines
drivers/net/phy/marvell.c +2 -4
-		 * delay between soft reset and loopback enablement.
-		 */
-		if (phydev->speed == SPEED_1000)
-			msleep(1000);
+		if (phydev->speed == SPEED_1000) {
+			err = phy_read_poll_timeout(phydev, MII_BMSR,

val, val & BMSR_LSTATUS,

Quoted 7 lines
Quoted 5 lines

Review static int m88e1510_loopback(struct · line 2022

Quoted 2 lines
(quoted diff) +2 -0
+						    PHY_LOOP_BACK_SLEEP,
+

PHY_LOOP_BACK_TIMEOUT, true);

Is this link with itself?

Its required cabled plug in, back to back connection.

Loopback should not require that. The whole point of loopback in the PHY is you can do it without needing a cable.

Quoted 5 lines
Quoted 2 lines

Have you tested this with the cable unplugged?

Yes we have and its expected to have the timeout. But the self-test required the link to be up first before it can be run.

So you get an ETIMEDOUT, and then skip the code which actually sets the LOOPBACK bit?

Please look at this again, and make it work without a cable.

Maybe you are addressing the wrong issue? Is the PHY actually performing loopback, but reporting the link is down? Maybe you need to fake a link up? Maybe you need the self test to not care about the link state, all it really needs is that packets get looped?

Andrew

#4 lore raw
Quoted 41 lines

-----Original Message-----
From: Andrew Lunn <[email protected]>
Sent: Tuesday, 30 August, 2022 8:17 PM
To: Jamaluddin, Aminuddin <[email protected]>
Cc: Heiner Kallweit <[email protected]>; Russell King <[email protected]>; David S . Miller <[email protected]>; Eric Dumazet <[email protected]>; Jakub Kicinski <[email protected]>; Paolo Abeni <[email protected]>; Ismail, Mohammad Athari <[email protected]>; [email protected]; linux- [email protected]; [email protected]; Tan, Tee Min <[email protected]>; Zulkifli, Muhammad Husaini <[email protected]>
Subject: Re: [PATCH net 1/1] net: phy: marvell: add link status check before enabling phy loopback

Quoted 14 lines
Quoted 14 lines

Review static int m88e1510_loopback(struct

Quoted 1 line
(quoted diff) +0 -0
@@ -2015,14 +2016,23 @@ static int m88e1510_loopback(struct

phy_device *phydev, bool enable)

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 4 lines
drivers/net/phy/marvell.c +0 -1
 		if (err < 0)
 			return err;
 
-		/* FIXME: Based on trial and error test, it seem 1G need to

have

drivers/net/phy/marvell.c static int m88e1510_loopback(struct phy_device *phydev, bool enable)

Quoted 6 lines
drivers/net/phy/marvell.c +2 -4
-		 * delay between soft reset and loopback enablement.
-		 */
-		if (phydev->speed == SPEED_1000)
-			msleep(1000);
+		if (phydev->speed == SPEED_1000) {
+			err = phy_read_poll_timeout(phydev, MII_BMSR,

val, val & BMSR_LSTATUS,

Quoted 7 lines
Quoted 5 lines

Review static int m88e1510_loopback(struct · line 2022

Quoted 2 lines
(quoted diff) +2 -0
+						    PHY_LOOP_BACK_SLEEP,
+

PHY_LOOP_BACK_TIMEOUT, true);

Is this link with itself?

Its required cabled plug in, back to back connection.

Loopback should not require that. The whole point of loopback in the PHY is you can do it without needing a cable.

Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Have you tested this with the cable unplugged?

Yes we have and its expected to have the timeout. But the self-test required the link to be up first before it can be run.

So you get an ETIMEDOUT, and then skip the code which actually sets the LOOPBACK bit?

If cable unplugged, test result will be displayed as 1. See comments below.

Quoted 2 lines

Please look at this again, and make it work without a cable.

Related to this the flow without cable, what we see in the codes during debugging. After the phy loopback bit was set.
The test will be run through this __stmmac_test_loopback()
https://elixir.bootlin.com/linux/v5.19.8/source/drivers/net/ethernet/stmicro/stmmac/stmmac_selftests.c#L320
Here, it will have another set of checking in dev_direct_xmit(), __dev_direct_xmit(). returning value 1(NET_XMIT_DROP)
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4288
Which means the interface is not available or the interface link status is not up. For this case the interface link status is not up.
Thus failing the phy loopback test.
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4296
Since we don't own this __stmmac_test_loopback(), we conclude the behaviour was as expected.

Quoted 5 lines

Maybe you are addressing the wrong issue? Is the PHY actually performing loopback, but reporting the link is down? Maybe you need to fake a link up? Maybe you need the self test to not care about the link state, all it really needs is that packets get looped?

When bit 14 was set, the link will be broken.
But before the self-test was triggered it requires link to be up as stated above comments.

Quoted 1 line

Amin

maintainer
#5 lore raw
Quoted 33 lines
Quoted 6 lines
Quoted 1 line

Its required cabled plug in, back to back connection.

Loopback should not require that. The whole point of loopback in the PHY is you can do it without needing a cable.

Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Have you tested this with the cable unplugged?

Yes we have and its expected to have the timeout. But the self-test required the link to be up first before it can be run.

So you get an ETIMEDOUT, and then skip the code which actually sets the LOOPBACK bit?

If cable unplugged, test result will be displayed as 1. See comments below.

Quoted 2 lines

Please look at this again, and make it work without a cable.

Related to this the flow without cable, what we see in the codes during debugging. After the phy loopback bit was set.
The test will be run through this __stmmac_test_loopback()
https://elixir.bootlin.com/linux/v5.19.8/source/drivers/net/ethernet/stmicro/stmmac/stmmac_selftests.c#L320
Here, it will have another set of checking in dev_direct_xmit(), __dev_direct_xmit(). returning value 1(NET_XMIT_DROP)
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4288
Which means the interface is not available or the interface link status is not up. For this case the interface link status is not up.
Thus failing the phy loopback test.
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4296
Since we don't own this __stmmac_test_loopback(), we conclude the behaviour was as expected.

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

Maybe you are addressing the wrong issue? Is the PHY actually performing loopback, but reporting the link is down? Maybe you need to fake a link up? Maybe you need the self test to not care about the link state, all it really needs is that packets get looped?

When bit 14 was set, the link will be broken.
But before the self-test was triggered it requires link to be up as stated above comments.

You have not said anything about my comment:

Quoted 1 line

Maybe you need to fake a link up?

My guess is, some PHYs are going to report link up when put into loopback. Others might not. For the Marvell PHY, it looks like you need to make marvell_read_status() return that the link is up if loopback is enabled.

Andrew

#6 lore raw
Quoted 63 lines

-----Original Message-----
From: Andrew Lunn <[email protected]>
Sent: Tuesday, 20 September, 2022 7:46 AM
To: Jamaluddin, Aminuddin <[email protected]>
Cc: Heiner Kallweit <[email protected]>; Russell King <[email protected]>; David S . Miller <[email protected]>; Eric Dumazet <[email protected]>; Jakub Kicinski <[email protected]>; Paolo Abeni <[email protected]>; Ismail, Mohammad Athari <[email protected]>; [email protected]; linux- [email protected]; [email protected]; Tan, Tee Min <[email protected]>; Zulkifli, Muhammad Husaini <[email protected]>
Subject: Re: [PATCH net 1/1] net: phy: marvell: add link status check before enabling phy loopback

Quoted 14 lines
Quoted 6 lines
Quoted 1 line

Its required cabled plug in, back to back connection.

Loopback should not require that. The whole point of loopback in the PHY is you can do it without needing a cable.

Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Have you tested this with the cable unplugged?

Yes we have and its expected to have the timeout. But the self-test required the link to be up first before it can be run.

So you get an ETIMEDOUT, and then skip the code which actually sets the LOOPBACK bit?

Quoted 6 lines

If cable unplugged, test result will be displayed as 1. See comments below.

Quoted 2 lines

Please look at this again, and make it work without a cable.

Related to this the flow without cable, what we see in the codes during

debugging.

Quoted 5 lines

After the phy loopback bit was set.
The test will be run through this __stmmac_test_loopback()
https://elixir.bootlin.com/linux/v5.19.8/source/drivers/net/ethernet/s
tmicro/stmmac/stmmac_selftests.c#L320
Here, it will have another set of checking in dev_direct_xmit(),

__dev_direct_xmit().

Quoted 3 lines

returning value 1(NET_XMIT_DROP)
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4288
Which means the interface is not available or the interface link status is not

up.

Quoted 4 lines

For this case the interface link status is not up.
Thus failing the phy loopback test.
https://elixir.bootlin.com/linux/v5.19.8/source/net/core/dev.c#L4296
Since we don't own this __stmmac_test_loopback(), we conclude the

behaviour was as expected.

Quoted 4 lines
Quoted 4 lines

Maybe you are addressing the wrong issue? Is the PHY actually performing loopback, but reporting the link is down? Maybe you need to

fake a link up?

Quoted 5 lines
Quoted 2 lines

Maybe you need the self test to not care about the link state, all it really needs is that packets get looped?

When bit 14 was set, the link will be broken.
But before the self-test was triggered it requires link to be up as stated

above comments.

Quoted 8 lines

You have not said anything about my comment:

Quoted 1 line

Maybe you need to fake a link up?

My guess is, some PHYs are going to report link up when put into loopback. Others might not. For the Marvell PHY, it looks like you need to make marvell_read_status() return that the link is up if loopback is enabled.

We able to do the PHY loopback test, after fake link up without cable plugged on as suggested above.
We will provide version 2 patch with minimum code changes without having the status link check.
Only need to increase the msleep(200) to msleep(1000) inside m88e1510_loopback() function.

Amin