Skip to contentSkip to conversation

drivers: net: phy: Ignore PHY ID 0 during PHY probing

PATCH New Alexandru Marginean 5 participants 16 messages Jun 19, 2019 – Jul 11, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
 drivers/net/phy/phy.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/net/phy/phy.c +9 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
 	while (phy_mask) {
 		int addr = ffs(phy_mask) - 1;
 		int r = get_phy_id(bus, addr, devad, &phy_id);
+
+		/* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs
+		 * that return all 0s for C22 reads (like Aquantia AQR112) and
+		 * there are C22 PHYs that return all 0s for C45 reads (like
+		 * Atheros AR8035).
+		 */
+		if (phy_id == 0)
+			return NULL;
+
 		/* If the PHY ID is mostly f's, we didn't find anything */
 		if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
 			is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 34 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
 drivers/net/phy/phy.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/net/phy/phy.c +9 -1
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
        while (phy_mask) {
                int addr = ffs(phy_mask) - 1;
                int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;
+
                /* If the PHY ID is mostly f's, we didn't find anything */
                if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
                        is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
--

2.17.1

Quoted 4 lines

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#2 lore raw

PATCH v2 1/1 drivers: net: phy: Use Aquantia driver for AQR112, AQR412

adds AQR112 and AQR412 to the list of supported PHYs using existing AQR code.

---

Changes in v2:
- Numerical ordering of structure definition and function calls

 drivers/net/phy/aquantia.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/net/phy/aquantia.c +28 -0
index 5c3298d612..465ec2d342 100644
@@ -461,6 +461,19 @@ struct phy_driver aqr107_driver =
 	.shutdown =en10g_shutdown,
 };
 
+struct phy_driver aqr112_driver =
+	.name =quantia AQR112",
+	.uid =3a1b660,
+	.mask =fffffff0,
+	.features =Y_10G_FEATURES,
+	.mmds =DIO_MMD_PMAPMD | MDIO_MMD_PCS |
+		 MDIO_MMD_PHYXS | MDIO_MMD_AN |
+		 MDIO_MMD_VEND1),
+	.config =quantia_config,
+	.startup =quantia_startup,
+	.shutdown =en10g_shutdown,
+};
+
 struct phy_driver aqr405_driver =
 	.name =quantia AQR405",
 	.uid =3a1b4b2,
@@ -474,6 +487,19 @@ struct phy_driver aqr405_driver =
 	.shutdown =en10g_shutdown,
 };
 
+struct phy_driver aqr412_driver =
+	.name =quantia AQR412",
+	.uid =3a1b710,
+	.mask =fffffff0,
+	.features =Y_10G_FEATURES,
+	.mmds =DIO_MMD_PMAPMD | MDIO_MMD_PCS |
+		 MDIO_MMD_PHYXS | MDIO_MMD_AN |
+		 MDIO_MMD_VEND1),
+	.config =quantia_config,
+	.startup =quantia_startup,
+	.shutdown =en10g_shutdown,
+};
+
 int phy_aquantia_init(void)
 {
 	phy_register(&aq1202_driver);
@@ -481,7 +507,9 @@ int phy_aquantia_init(void)
 	phy_register(&aqr105_driver);
 	phy_register(&aqr106_driver);
 	phy_register(&aqr107_driver);
+	phy_register(&aqr112_driver);
 	phy_register(&aqr405_driver);
+	phy_register(&aqr412_driver);
 
 	return 0;
 }
Signature · 1 line
2.17.1
#3 lore raw

PATCH v2 1/1 drivers: net: phy: Use Aquantia driver for AQR112, AQR412

Quoted 68 lines · Alexandru Marginean

adds AQR112 and AQR412 to the list of supported PHYs using existing AQR code.

---

Changes in v2:
- Numerical ordering of structure definition and function calls

 drivers/net/phy/aquantia.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/net/phy/aquantia.c +28 -1
index 5c3298d612..465ec2d342 100644
@@ -461,6 +461,19 @@ struct phy_driver aqr107_driver =
        .shutdown =en10g_shutdown,
 };
 
+struct phy_driver aqr112_driver =
+       .name =quantia AQR112",
+       .uid =3a1b660,
+       .mask =fffffff0,
+       .features =Y_10G_FEATURES,
+       .mmds =DIO_MMD_PMAPMD | MDIO_MMD_PCS |
+                MDIO_MMD_PHYXS | MDIO_MMD_AN |
+                MDIO_MMD_VEND1),
+       .config =quantia_config,
+       .startup =quantia_startup,
+       .shutdown =en10g_shutdown,
+};
+
 struct phy_driver aqr405_driver =
        .name =quantia AQR405",
        .uid =3a1b4b2,
@@ -474,6 +487,19 @@ struct phy_driver aqr405_driver =
        .shutdown =en10g_shutdown,
 };
 
+struct phy_driver aqr412_driver =
+       .name =quantia AQR412",
+       .uid =3a1b710,
+       .mask =fffffff0,
+       .features =Y_10G_FEATURES,
+       .mmds =DIO_MMD_PMAPMD | MDIO_MMD_PCS |
+                MDIO_MMD_PHYXS | MDIO_MMD_AN |
+                MDIO_MMD_VEND1),
+       .config =quantia_config,
+       .startup =quantia_startup,
+       .shutdown =en10g_shutdown,
+};
+
 int phy_aquantia_init(void)
 {
        phy_register(&aq1202_driver);
@@ -481,7 +507,9 @@ int phy_aquantia_init(void)
        phy_register(&aqr105_driver);
        phy_register(&aqr106_driver);
        phy_register(&aqr107_driver);
+       phy_register(&aqr112_driver);
        phy_register(&aqr405_driver);
+       phy_register(&aqr412_driver);
 
        return 0;
 }
--
Quoted 2 lines

2.17.1

#4 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

In reply to Alexandru Marginean

Hi Alex,

Quoted 16 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
 drivers/net/phy/phy.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/net/phy/phy.c +0 -0
index c1c1af9abd..7ccbc4d9da 100644

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 5 lines
drivers/net/phy/phy.c +2 -0
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
        while (phy_mask) {
                int addr = ffs(phy_mask) - 1;
                int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 6 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus,

Quoted 5 lines
drivers/net/phy/phy.c +1 -1
+
                /* If the PHY ID is mostly f's, we didn't find anything */
                if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
                        is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
--

Regards,
Bin

#5 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
  drivers/net/phy/phy.c | 9 +++++++++
  1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
         while (phy_mask) {
                 int addr = ffs(phy_mask) - 1;
                 int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.
If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

What do you think?

Thank you!
Alex

Quoted 14 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus,

Quoted 6 lines
drivers/net/phy/phy.c +1 -1
+
                 /* If the PHY ID is mostly f's, we didn't find anything */
                 if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
                         is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
--

Regards,
Bin
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#6 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
  drivers/net/phy/phy.c | 9 +++++++++
  1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
         while (phy_mask) {
                 int addr = ffs(phy_mask) - 1;
                 int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Quoted 2 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

Quoted 8 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

What do you think?

Regards,
Bin

#7 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
   drivers/net/phy/phy.c | 9 +++++++++
   1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
          while (phy_mask) {
                  int addr = ffs(phy_mask) - 1;
                  int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Thank you!
Alex

Quoted 12 lines
Quoted 8 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

What do you think?

Regards,
Bin

#8 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
   drivers/net/phy/phy.c | 9 +++++++++
   1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
          while (phy_mask) {
                  int addr = ffs(phy_mask) - 1;
                  int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Quoted 21 lines

Thank you!
Alex

Quoted 7 lines
Quoted 7 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

Quoted 5 lines
Quoted 1 line

What do you think?

Regards,
Bin

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#9 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Hi Joe,

Quoted 101 lines · Joe Hershberger
Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
    drivers/net/phy/phy.c | 9 +++++++++
    1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
           while (phy_mask) {
                   int addr = ffs(phy_mask) - 1;
                   int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Quoted 8 lines

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Assuming there's just one PHY on the bus that's scanned, and it's a C45 PHY, scanning should actually work (that case works with both continue and return NULL in fact).

create_phy_by_mask is going to read PHY_ID 0 over C22, it can continue and not find anything else using C22 reads, finally return NULL. get_phy_device_by_mask will call again with the next devad. Assuming the PHY replies to devad 1 or one of the other ones used in get_phy_device_by_mask, the PHY will be probed as a C45 PHY.

Things aren't as good if there is a bus with multiple PHYs, especially if they are mixed, C22 with C45. In that case relying on scanning looks like a bad idea though.

Alex

Quoted 19 lines
Quoted 19 lines
Quoted 9 lines
Quoted 9 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

Quoted 5 lines
Quoted 1 line

What do you think?

Regards,
Bin

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#10 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 124 lines · Alex Marginean

Hi Joe,

Quoted 101 lines · Joe Hershberger
Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
    drivers/net/phy/phy.c | 9 +++++++++
    1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
           while (phy_mask) {
                   int addr = ffs(phy_mask) - 1;
                   int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Quoted 8 lines

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Assuming there's just one PHY on the bus that's scanned, and it's a C45 PHY, scanning should actually work (that case works with both continue and return NULL in fact).

create_phy_by_mask is going to read PHY_ID 0 over C22, it can continue and not find anything else using C22 reads, finally return NULL. get_phy_device_by_mask will call again with the next devad. Assuming the PHY replies to devad 1 or one of the other ones used in get_phy_device_by_mask, the PHY will be probed as a C45 PHY.

Quoted 3 lines

Things aren't as good if there is a bus with multiple PHYs, especially if they are mixed, C22 with C45. In that case relying on scanning looks like a bad idea though.

I agree... if you have more than one phy, on the same MDIO, you had better have them mapped to MACs explicitly. I'm more curious if there are use-cases where you would want to ignore a C45 phy. What kind of performance hit would we take for continuing to scan, just so that we can warn that the DT should list the phy addrs?

Quoted 27 lines

Alex

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

What do you think?

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

Regards,
Bin

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#11 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 129 lines · Joe Hershberger
Quoted 124 lines · Alex Marginean

Hi Joe,

Quoted 101 lines · Joe Hershberger
Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
     drivers/net/phy/phy.c | 9 +++++++++
     1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
            while (phy_mask) {
                    int addr = ffs(phy_mask) - 1;
                    int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Quoted 8 lines

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Assuming there's just one PHY on the bus that's scanned, and it's a C45 PHY, scanning should actually work (that case works with both continue and return NULL in fact).

create_phy_by_mask is going to read PHY_ID 0 over C22, it can continue and not find anything else using C22 reads, finally return NULL. get_phy_device_by_mask will call again with the next devad. Assuming the PHY replies to devad 1 or one of the other ones used in get_phy_device_by_mask, the PHY will be probed as a C45 PHY.

Quoted 3 lines

Things aren't as good if there is a bus with multiple PHYs, especially if they are mixed, C22 with C45. In that case relying on scanning looks like a bad idea though.

Quoted 5 lines

I agree... if you have more than one phy, on the same MDIO, you had better have them mapped to MACs explicitly. I'm more curious if there are use-cases where you would want to ignore a C45 phy. What kind of performance hit would we take for continuing to scan, just so that we can warn that the DT should list the phy addrs?

Top of my head, reading all 32 addresses could go up to a few milliseconds. This is really only useful for development set-ups really, for devices in the field it doesn't feel right to scan the entire bus if it happens that the only PHY on that bus is at a low addr. Not to mention that no one would see a warning anyway.
So maybe include this as a debug option, or rather just document that phy_find_by_mask should only be called if there is a single PHY on the bus, otherwise the result is not guaranteed.

Quoted 28 lines
Quoted 28 lines

Alex

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

What do you think?

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

Regards,
Bin

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#12 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 137 lines · Alex Marginean
Quoted 129 lines · Joe Hershberger
Quoted 124 lines · Alex Marginean

Hi Joe,

Quoted 101 lines · Joe Hershberger
Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
     drivers/net/phy/phy.c | 9 +++++++++
     1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
            while (phy_mask) {
                    int addr = ffs(phy_mask) - 1;
                    int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Quoted 8 lines

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Assuming there's just one PHY on the bus that's scanned, and it's a C45 PHY, scanning should actually work (that case works with both continue and return NULL in fact).

create_phy_by_mask is going to read PHY_ID 0 over C22, it can continue and not find anything else using C22 reads, finally return NULL. get_phy_device_by_mask will call again with the next devad. Assuming the PHY replies to devad 1 or one of the other ones used in get_phy_device_by_mask, the PHY will be probed as a C45 PHY.

Quoted 3 lines

Things aren't as good if there is a bus with multiple PHYs, especially if they are mixed, C22 with C45. In that case relying on scanning looks like a bad idea though.

Quoted 5 lines

I agree... if you have more than one phy, on the same MDIO, you had better have them mapped to MACs explicitly. I'm more curious if there are use-cases where you would want to ignore a C45 phy. What kind of performance hit would we take for continuing to scan, just so that we can warn that the DT should list the phy addrs?

Quoted 8 lines

Top of my head, reading all 32 addresses could go up to a few milliseconds. This is really only useful for development set-ups really, for devices in the field it doesn't feel right to scan the entire bus if it happens that the only PHY on that bus is at a low addr. Not to mention that no one would see a warning anyway.
So maybe include this as a debug option, or rather just document that phy_find_by_mask should only be called if there is a single PHY on the bus, otherwise the result is not guaranteed.

I agree that it doesn't make sense in production. I'm not sure there's enough value in dev to add this feature. Adding a little documentation is probably about the right level of work.

Quoted 33 lines
Quoted 24 lines
Quoted 24 lines

Alex

Quoted 19 lines
Quoted 19 lines
Quoted 9 lines
Quoted 9 lines

smallest addr. This change shouldn't break that, assuming PHY_ID is != 0 for at least one devad used. Or we could keep the current devad and continue scanning for the next addr, continue instead of return would achieve that. I don't really have a strong preference for either, the message for developers should be to avoid ending up in this situation altogether.

Quoted 5 lines
Quoted 1 line

What do you think?

Regards,
Bin

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

Quoted 4 lines
Quoted 4 lines

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#13 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Quoted 147 lines · Joe Hershberger
Quoted 137 lines · Alex Marginean
Quoted 129 lines · Joe Hershberger
Quoted 124 lines · Alex Marginean

Hi Joe,

Quoted 101 lines · Joe Hershberger
Quoted 93 lines · Alex Marginean

Hi Bin,

Quoted 58 lines · Bin Meng

Hi Alex,

Quoted 47 lines · Alex Marginean
Quoted 15 lines · Bin Meng

Hi Alex,

Quoted 11 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
      drivers/net/phy/phy.c | 9 +++++++++
      1 file changed, 9 insertions(+)
Quoted 12 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 10 lines
drivers/net/phy/phy.c +2 -0
index c1c1af9abd..7ccbc4d9da 100644
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
             while (phy_mask) {
                     int addr = ffs(phy_mask) - 1;
                     int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

nits: the multi-line comment block format is wrong

You're right, I'll fix that.

Quoted 9 lines

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 7 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

Should this be "continue"?

In case there are C45 and C22 PHYs mixed on the same bus and they are all flagged in phy_mask? In general this function shouldn't end up dealing with multiple PHYs, if it does then it's possible the result won't the the right one.

Quoted 7 lines

If create_phy_by_mask is called from get_phy_device, then we're only looking for one PHY and if that reads PHY_ID 0 we can just assume we're not using the correct devad.

create_phy_by_mask can also be called from phy_connect (or some other place) with phy_mask = 0xffffffff. The assumption in that case is that there is one PHY on the given MDIO bus.

Quoted 6 lines

Yes, this is the user scenario that concerns me. But on a shared MDIO bus, there are multiple PHYs. I remember lots of old Freescale PowerPC boards did this way. For example, there are multiple eTSEC in the SoC, and each eTSEC claims to have one MDIO controller, however Freescale chose to wire all PHYs on a single MDIO bus which usually is eTSEC0 (the first eTSEC).

Virtually all Freescale networking SoCs are like that, each MAC has its own MDIO controller but those are in fact connected to the SoC internal PHYs not to the external PHYs on the board. These SoCs have one or two MDIOs dedicated to the external PHYs. If the SoC supports 10G interfaces then the SoC typically has two external MDIOs, the intent being that one is used for C22 and the other for C45 PHYs. Of course MDIO buses with multiple PHYs have to work. My point is that one should not end up in this function with a phy_mask with multiple bits set if the MDIO bus has multiple PHYs connected to it, in that set-up the caller should know the PHY address up front and set only one bit in phy_mask.

Quoted 6 lines
Quoted 3 lines

If there are several PHYs then we're going into a gray area, the result isn't explicitly defined. We could try to probe the PHY with the

I suspect this function is not being used widely hence not exposing any known issues?

At least for qoriq and layerscape Freescale SoCs the PHY address is known up front before calling this function, precisely because the MDIO bus is shared, but I'm guessing other SoCs do rely on scanning to get to the PHYs.

Quoted 6 lines

Joe, do you have a preference between return and continue when we hit a PHY_ID == 0? Are you OK with continue?

I wonder if it would be useful to scan all IDs in this function, instead of returning on 1st hit, just to issue a warning if multiple addresses flagged in phy_mask return valid PHY IDs. Any thoughts?

Quoted 8 lines

Scanning all seems like a bit of an independent question. I think the return vs continue decision at a higher level is a decision about whether a phy scan should include or ignore C45 phys, correct? Scan all seems like it applies to all included phys, C45 being included or not based on the former decision.

Is there a use case anyone is aware of where C45 phys are expected to be scanned for?

Assuming there's just one PHY on the bus that's scanned, and it's a C45 PHY, scanning should actually work (that case works with both continue and return NULL in fact).

create_phy_by_mask is going to read PHY_ID 0 over C22, it can continue and not find anything else using C22 reads, finally return NULL. get_phy_device_by_mask will call again with the next devad. Assuming the PHY replies to devad 1 or one of the other ones used in get_phy_device_by_mask, the PHY will be probed as a C45 PHY.

Quoted 3 lines

Things aren't as good if there is a bus with multiple PHYs, especially if they are mixed, C22 with C45. In that case relying on scanning looks like a bad idea though.

Quoted 5 lines

I agree... if you have more than one phy, on the same MDIO, you had better have them mapped to MACs explicitly. I'm more curious if there are use-cases where you would want to ignore a C45 phy. What kind of performance hit would we take for continuing to scan, just so that we can warn that the DT should list the phy addrs?

Quoted 8 lines

Top of my head, reading all 32 addresses could go up to a few milliseconds. This is really only useful for development set-ups really, for devices in the field it doesn't feel right to scan the entire bus if it happens that the only PHY on that bus is at a low addr. Not to mention that no one would see a warning anyway.
So maybe include this as a debug option, or rather just document that phy_find_by_mask should only be called if there is a single PHY on the bus, otherwise the result is not guaranteed.

Quoted 3 lines

I agree that it doesn't make sense in production. I'm not sure there's enough value in dev to add this feature. Adding a little documentation is probably about the right level of work.

Sent as a separate patch with comments for PHY APIs.

Thank you!
Alex

#14 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

In reply to Alexandru Marginean
Quoted 16 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
 drivers/net/phy/phy.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/net/phy/phy.c +0 -0
index c1c1af9abd..7ccbc4d9da 100644

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 5 lines
drivers/net/phy/phy.c +2 -0
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
        while (phy_mask) {
                int addr = ffs(phy_mask) - 1;
                int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

Agreed on the format.

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 6 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;

As for continue or return, this should probably check that phy_id == 0 and r == 0, right?

Quoted 11 lines
drivers/net/phy/phy.c +1 -1
+
                /* If the PHY ID is mostly f's, we didn't find anything */
                if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
                        is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
--

2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#15 lore raw

PATCH 1/1 drivers: net: phy: Ignore PHY ID 0 during PHY probing

Hi Joe,

Quoted 33 lines · Joe Hershberger
Quoted 16 lines · Alexandru Marginean

Current code fails to probe some C45 PHYs that also respond to C22 reads. This is the case for PHYs like Aquantia AQR112, Marvell 88X2242 (as previously posted on the u-boot list).
If the PHY ID reads all 0s just ignore it and try the next devad.

---
  drivers/net/phy/phy.c | 9 +++++++++
  1 file changed, 9 insertions(+)
drivers/net/phy/phy.c +0 -0
index c1c1af9abd..7ccbc4d9da 100644

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 731

Quoted 5 lines
drivers/net/phy/phy.c +2 -0
@@ -727,6 +727,15 @@ static struct phy_device *create_phy_by_mask(struct mii_dev *bus,
         while (phy_mask) {
                 int addr = ffs(phy_mask) - 1;
                 int r = get_phy_id(bus, addr, devad, &phy_id);
+
+               /* If the PHY ID is flat 0 we ignore it.  There are C45 PHYs

Agreed on the format.

drivers/net/phy/phy.c static struct phy_device *create_phy_by_mask(struct mii_dev *bus, · line 737

Quoted 6 lines
drivers/net/phy/phy.c +6 -0
+                * that return all 0s for C22 reads (like Aquantia AQR112) and
+                * there are C22 PHYs that return all 0s for C45 reads (like
+                * Atheros AR8035).
+                */
+               if (phy_id == 0)
+                       return NULL;
Quoted 2 lines

As for continue or return, this should probably check that phy_id == 0 and r == 0, right?

Right, in case the read over MDIO returns an error code for the missing PHY we want to continue the loop. I'll fix this.

Thank you!
Alex

Quoted 17 lines
Quoted 8 lines
drivers/net/phy/phy.c +1 -1
+
                 /* If the PHY ID is mostly f's, we didn't find anything */
                 if (r == 0 && (phy_id & 0x1fffffff) != 0x1fffffff) {
                         is_c45 = (devad == MDIO_DEVAD_NONE) ? false : true;
--

2.17.1

Quoted 4 lines

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot