The USB2 PHY mode handling on Tegra210 incorrectly relied on
regulator_is_enabled() when determining whether the VBUS supply should
be disabled during role changes. This is because regulator_is_enabled()
reports exactly what is states and not if there is an unbalanced number
of calls between regulator_enable() and regulator_disable(). For
example, regulator_is_enabled() always reports true on a fixed-regulator
with no enable gpio, which is the case on the Pixel C.

This then leads to the PHY driver wrongfully calling regulator_disable()
when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the driver
did not previously call the corresponding regulator_enable().

Fix this by keeping track of the current role and updating the logic to
disable the regulator only when the previous role was USB_ROLE_HOST.

While at it fix a small typo in a comment.

Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
---
 drivers/phy/tegra/xusb-tegra210.c | 5 +++--
 drivers/phy/tegra/xusb.h          | 1 +
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/xusb-tegra210.c
index 3409924498e9..63ad57d95514 100644
--- a/drivers/phy/tegra/xusb-tegra210.c
+++ b/drivers/phy/tegra/xusb-tegra210.c
@@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
 			/*
 			 * When port is peripheral only or role transitions to
 			 * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
-			 * be enabled.
+			 * enabled.
 			 */
-			if (regulator_is_enabled(port->supply))
+			if (port->role == USB_ROLE_HOST)
 				regulator_disable(port->supply);
 
 			tegra210_xusb_padctl_id_override(padctl, false);
@@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
 		}
 	}
 
+	port->role = submode;
 	mutex_unlock(&padctl->lock);
 
 	return err;
diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
index d2b5f9565132..273af147dfd3 100644
--- a/drivers/phy/tegra/xusb.h
+++ b/drivers/phy/tegra/xusb.h
@@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
 	enum usb_dr_mode mode;
 	bool internal;
 	int usb3_port_fake;
+	enum usb_role role;
 };
 
 static inline struct tegra_xusb_usb2_port *

-- 
2.52.0

On 04/12/2025 21:27, Diogo Ivo wrote:
> The USB2 PHY mode handling on Tegra210 incorrectly relied on
> regulator_is_enabled() when determining whether the VBUS supply should
> be disabled during role changes. This is because regulator_is_enabled()
> reports exactly what is states and not if there is an unbalanced number
> of calls between regulator_enable() and regulator_disable(). For
> example, regulator_is_enabled() always reports true on a fixed-regulator
> with no enable gpio, which is the case on the Pixel C.
> 
> This then leads to the PHY driver wrongfully calling regulator_disable()
> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the driver
> did not previously call the corresponding regulator_enable().
> 
> Fix this by keeping track of the current role and updating the logic to
> disable the regulator only when the previous role was USB_ROLE_HOST.
> 
> While at it fix a small typo in a comment.
> 
> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
> ---
>   drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>   drivers/phy/tegra/xusb.h          | 1 +
>   2 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/xusb-tegra210.c
> index 3409924498e9..63ad57d95514 100644
> --- a/drivers/phy/tegra/xusb-tegra210.c
> +++ b/drivers/phy/tegra/xusb-tegra210.c
> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>   			/*
>   			 * When port is peripheral only or role transitions to
>   			 * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
> -			 * be enabled.
> +			 * enabled.
>   			 */
> -			if (regulator_is_enabled(port->supply))
> +			if (port->role == USB_ROLE_HOST)
>   				regulator_disable(port->supply);
>   
>   			tegra210_xusb_padctl_id_override(padctl, false);
> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>   		}
>   	}
>   
> +	port->role = submode;
>   	mutex_unlock(&padctl->lock);
>   
>   	return err;
> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
> index d2b5f9565132..273af147dfd3 100644
> --- a/drivers/phy/tegra/xusb.h
> +++ b/drivers/phy/tegra/xusb.h
> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>   	enum usb_dr_mode mode;
>   	bool internal;
>   	int usb3_port_fake;
> +	enum usb_role role;
>   };


A similar fix was made to the Tegra186 code by commit cefc1caee9dd 
("phy: tegra: xusb: Fix unbalanced regulator disable in UTMI PHY mode"). 
Although the above looks simpler, I am wondering if we should make a 
similar change to the Tegra210 code so that they both are implemented in 
the same way?

Jon
-- 
nvpublic

On 1/13/26 12:01, Jon Hunter wrote:
> 
> On 04/12/2025 21:27, Diogo Ivo wrote:
>> The USB2 PHY mode handling on Tegra210 incorrectly relied on
>> regulator_is_enabled() when determining whether the VBUS supply should
>> be disabled during role changes. This is because regulator_is_enabled()
>> reports exactly what is states and not if there is an unbalanced number
>> of calls between regulator_enable() and regulator_disable(). For
>> example, regulator_is_enabled() always reports true on a fixed-regulator
>> with no enable gpio, which is the case on the Pixel C.
>>
>> This then leads to the PHY driver wrongfully calling regulator_disable()
>> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the driver
>> did not previously call the corresponding regulator_enable().
>>
>> Fix this by keeping track of the current role and updating the logic to
>> disable the regulator only when the previous role was USB_ROLE_HOST.
>>
>> While at it fix a small typo in a comment.
>>
>> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
>> ---
>>   drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>>   drivers/phy/tegra/xusb.h          | 1 +
>>   2 files changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/ 
>> xusb-tegra210.c
>> index 3409924498e9..63ad57d95514 100644
>> --- a/drivers/phy/tegra/xusb-tegra210.c
>> +++ b/drivers/phy/tegra/xusb-tegra210.c
>> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct phy 
>> *phy, enum phy_mode mode,
>>               /*
>>                * When port is peripheral only or role transitions to
>>                * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
>> -             * be enabled.
>> +             * enabled.
>>                */
>> -            if (regulator_is_enabled(port->supply))
>> +            if (port->role == USB_ROLE_HOST)
>>                   regulator_disable(port->supply);
>>               tegra210_xusb_padctl_id_override(padctl, false);
>> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct phy 
>> *phy, enum phy_mode mode,
>>           }
>>       }
>> +    port->role = submode;
>>       mutex_unlock(&padctl->lock);
>>       return err;
>> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
>> index d2b5f9565132..273af147dfd3 100644
>> --- a/drivers/phy/tegra/xusb.h
>> +++ b/drivers/phy/tegra/xusb.h
>> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>>       enum usb_dr_mode mode;
>>       bool internal;
>>       int usb3_port_fake;
>> +    enum usb_role role;
>>   };
> 
> 
> A similar fix was made to the Tegra186 code by commit cefc1caee9dd 
> ("phy: tegra: xusb: Fix unbalanced regulator disable in UTMI PHY mode"). 
> Although the above looks simpler, I am wondering if we should make a 
> similar change to the Tegra210 code so that they both are implemented in 
> the same way?

Looking at cefc1caee9dd my approach leads to less changes but I do agree
that standardization benefits us here. However in that case I think we
can take it a step further and actually just have a single function
tegra_xusb_padctl_id_override() (and likewise for vbus_override() and
set_mode()) since they all seem to do the same thing in both platforms.

If not I can simply make this patch look like cefc1caee9dd. Let me know
what you think!

Diogo

> Jon

On 13/01/2026 13:59, Diogo Ivo wrote:
> 
> 
> On 1/13/26 12:01, Jon Hunter wrote:
>>
>> On 04/12/2025 21:27, Diogo Ivo wrote:
>>> The USB2 PHY mode handling on Tegra210 incorrectly relied on
>>> regulator_is_enabled() when determining whether the VBUS supply should
>>> be disabled during role changes. This is because regulator_is_enabled()
>>> reports exactly what is states and not if there is an unbalanced number
>>> of calls between regulator_enable() and regulator_disable(). For
>>> example, regulator_is_enabled() always reports true on a fixed-regulator
>>> with no enable gpio, which is the case on the Pixel C.
>>>
>>> This then leads to the PHY driver wrongfully calling regulator_disable()
>>> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the 
>>> driver
>>> did not previously call the corresponding regulator_enable().
>>>
>>> Fix this by keeping track of the current role and updating the logic to
>>> disable the regulator only when the previous role was USB_ROLE_HOST.
>>>
>>> While at it fix a small typo in a comment.
>>>
>>> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
>>> ---
>>>   drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>>>   drivers/phy/tegra/xusb.h          | 1 +
>>>   2 files changed, 4 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/ 
>>> xusb-tegra210.c
>>> index 3409924498e9..63ad57d95514 100644
>>> --- a/drivers/phy/tegra/xusb-tegra210.c
>>> +++ b/drivers/phy/tegra/xusb-tegra210.c
>>> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct 
>>> phy *phy, enum phy_mode mode,
>>>               /*
>>>                * When port is peripheral only or role transitions to
>>>                * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
>>> -             * be enabled.
>>> +             * enabled.
>>>                */
>>> -            if (regulator_is_enabled(port->supply))
>>> +            if (port->role == USB_ROLE_HOST)
>>>                   regulator_disable(port->supply);
>>>               tegra210_xusb_padctl_id_override(padctl, false);
>>> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct 
>>> phy *phy, enum phy_mode mode,
>>>           }
>>>       }
>>> +    port->role = submode;
>>>       mutex_unlock(&padctl->lock);
>>>       return err;
>>> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
>>> index d2b5f9565132..273af147dfd3 100644
>>> --- a/drivers/phy/tegra/xusb.h
>>> +++ b/drivers/phy/tegra/xusb.h
>>> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>>>       enum usb_dr_mode mode;
>>>       bool internal;
>>>       int usb3_port_fake;
>>> +    enum usb_role role;
>>>   };
>>
>>
>> A similar fix was made to the Tegra186 code by commit cefc1caee9dd 
>> ("phy: tegra: xusb: Fix unbalanced regulator disable in UTMI PHY 
>> mode"). Although the above looks simpler, I am wondering if we should 
>> make a similar change to the Tegra210 code so that they both are 
>> implemented in the same way?
> 
> Looking at cefc1caee9dd my approach leads to less changes but I do agree
> that standardization benefits us here. However in that case I think we
> can take it a step further and actually just have a single function
> tegra_xusb_padctl_id_override() (and likewise for vbus_override() and
> set_mode()) since they all seem to do the same thing in both platforms.

Yes I think that would be fine. I can't say I have looked at that in 
detail but that would seem like the logical way to go.

Jon

-- 
nvpublic

On 1/13/26 14:42, Jon Hunter wrote:
> 
> On 13/01/2026 13:59, Diogo Ivo wrote:
>>
>>
>> On 1/13/26 12:01, Jon Hunter wrote:
>>>
>>> On 04/12/2025 21:27, Diogo Ivo wrote:
>>>> The USB2 PHY mode handling on Tegra210 incorrectly relied on
>>>> regulator_is_enabled() when determining whether the VBUS supply should
>>>> be disabled during role changes. This is because regulator_is_enabled()
>>>> reports exactly what is states and not if there is an unbalanced number
>>>> of calls between regulator_enable() and regulator_disable(). For
>>>> example, regulator_is_enabled() always reports true on a fixed- 
>>>> regulator
>>>> with no enable gpio, which is the case on the Pixel C.
>>>>
>>>> This then leads to the PHY driver wrongfully calling 
>>>> regulator_disable()
>>>> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the 
>>>> driver
>>>> did not previously call the corresponding regulator_enable().
>>>>
>>>> Fix this by keeping track of the current role and updating the logic to
>>>> disable the regulator only when the previous role was USB_ROLE_HOST.
>>>>
>>>> While at it fix a small typo in a comment.
>>>>
>>>> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
>>>> ---
>>>>   drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>>>>   drivers/phy/tegra/xusb.h          | 1 +
>>>>   2 files changed, 4 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/ 
>>>> xusb-tegra210.c
>>>> index 3409924498e9..63ad57d95514 100644
>>>> --- a/drivers/phy/tegra/xusb-tegra210.c
>>>> +++ b/drivers/phy/tegra/xusb-tegra210.c
>>>> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct 
>>>> phy *phy, enum phy_mode mode,
>>>>               /*
>>>>                * When port is peripheral only or role transitions to
>>>>                * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
>>>> -             * be enabled.
>>>> +             * enabled.
>>>>                */
>>>> -            if (regulator_is_enabled(port->supply))
>>>> +            if (port->role == USB_ROLE_HOST)
>>>>                   regulator_disable(port->supply);
>>>>               tegra210_xusb_padctl_id_override(padctl, false);
>>>> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct 
>>>> phy *phy, enum phy_mode mode,
>>>>           }
>>>>       }
>>>> +    port->role = submode;
>>>>       mutex_unlock(&padctl->lock);
>>>>       return err;
>>>> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
>>>> index d2b5f9565132..273af147dfd3 100644
>>>> --- a/drivers/phy/tegra/xusb.h
>>>> +++ b/drivers/phy/tegra/xusb.h
>>>> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>>>>       enum usb_dr_mode mode;
>>>>       bool internal;
>>>>       int usb3_port_fake;
>>>> +    enum usb_role role;
>>>>   };
>>>
>>>
>>> A similar fix was made to the Tegra186 code by commit cefc1caee9dd 
>>> ("phy: tegra: xusb: Fix unbalanced regulator disable in UTMI PHY 
>>> mode"). Although the above looks simpler, I am wondering if we should 
>>> make a similar change to the Tegra210 code so that they both are 
>>> implemented in the same way?
>>
>> Looking at cefc1caee9dd my approach leads to less changes but I do agree
>> that standardization benefits us here. However in that case I think we
>> can take it a step further and actually just have a single function
>> tegra_xusb_padctl_id_override() (and likewise for vbus_override() and
>> set_mode()) since they all seem to do the same thing in both platforms.
> 
> Yes I think that would be fine. I can't say I have looked at that in 
> detail but that would seem like the logical way to go.

Ok, then I'll do just that.

> Jon

On 04-12-25, 21:27, Diogo Ivo wrote:
> The USB2 PHY mode handling on Tegra210 incorrectly relied on
> regulator_is_enabled() when determining whether the VBUS supply should
> be disabled during role changes. This is because regulator_is_enabled()
> reports exactly what is states and not if there is an unbalanced number
> of calls between regulator_enable() and regulator_disable(). For
> example, regulator_is_enabled() always reports true on a fixed-regulator
> with no enable gpio, which is the case on the Pixel C.
> 
> This then leads to the PHY driver wrongfully calling regulator_disable()
> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the driver
> did not previously call the corresponding regulator_enable().
> 
> Fix this by keeping track of the current role and updating the logic to
> disable the regulator only when the previous role was USB_ROLE_HOST.
> 
> While at it fix a small typo in a comment.

Never mix a patch with something else please. More imp if it is fix
which will go to rcX. Please send a different patch for typo

> 
> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
> ---
>  drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>  drivers/phy/tegra/xusb.h          | 1 +
>  2 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/xusb-tegra210.c
> index 3409924498e9..63ad57d95514 100644
> --- a/drivers/phy/tegra/xusb-tegra210.c
> +++ b/drivers/phy/tegra/xusb-tegra210.c
> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>  			/*
>  			 * When port is peripheral only or role transitions to
>  			 * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
> -			 * be enabled.
> +			 * enabled.
>  			 */
> -			if (regulator_is_enabled(port->supply))
> +			if (port->role == USB_ROLE_HOST)
>  				regulator_disable(port->supply);
>  
>  			tegra210_xusb_padctl_id_override(padctl, false);
> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>  		}
>  	}
>  
> +	port->role = submode;
>  	mutex_unlock(&padctl->lock);
>  
>  	return err;
> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
> index d2b5f9565132..273af147dfd3 100644
> --- a/drivers/phy/tegra/xusb.h
> +++ b/drivers/phy/tegra/xusb.h
> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>  	enum usb_dr_mode mode;
>  	bool internal;
>  	int usb3_port_fake;
> +	enum usb_role role;
>  };

Jonathan can we get some t-b for these two patches

-- 
~Vinod

On 12/24/25 07:27, Vinod Koul wrote:
> On 04-12-25, 21:27, Diogo Ivo wrote:
>> The USB2 PHY mode handling on Tegra210 incorrectly relied on
>> regulator_is_enabled() when determining whether the VBUS supply should
>> be disabled during role changes. This is because regulator_is_enabled()
>> reports exactly what is states and not if there is an unbalanced number
>> of calls between regulator_enable() and regulator_disable(). For
>> example, regulator_is_enabled() always reports true on a fixed-regulator
>> with no enable gpio, which is the case on the Pixel C.
>>
>> This then leads to the PHY driver wrongfully calling regulator_disable()
>> when transitioning from USB_ROLE_DEVICE to USB_ROLE_NONE since the driver
>> did not previously call the corresponding regulator_enable().
>>
>> Fix this by keeping track of the current role and updating the logic to
>> disable the regulator only when the previous role was USB_ROLE_HOST.
>>
>> While at it fix a small typo in a comment.
> 
> Never mix a patch with something else please. More imp if it is fix
> which will go to rcX. Please send a different patch for typo

Ok, will split for v2.

>> Signed-off-by: Diogo Ivo <diogo.ivo@tecnico.ulisboa.pt>
>> ---
>>   drivers/phy/tegra/xusb-tegra210.c | 5 +++--
>>   drivers/phy/tegra/xusb.h          | 1 +
>>   2 files changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/phy/tegra/xusb-tegra210.c b/drivers/phy/tegra/xusb-tegra210.c
>> index 3409924498e9..63ad57d95514 100644
>> --- a/drivers/phy/tegra/xusb-tegra210.c
>> +++ b/drivers/phy/tegra/xusb-tegra210.c
>> @@ -1934,9 +1934,9 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>>   			/*
>>   			 * When port is peripheral only or role transitions to
>>   			 * USB_ROLE_NONE from USB_ROLE_DEVICE, regulator is not
>> -			 * be enabled.
>> +			 * enabled.
>>   			 */
>> -			if (regulator_is_enabled(port->supply))
>> +			if (port->role == USB_ROLE_HOST)
>>   				regulator_disable(port->supply);
>>   
>>   			tegra210_xusb_padctl_id_override(padctl, false);
>> @@ -1944,6 +1944,7 @@ static int tegra210_usb2_phy_set_mode(struct phy *phy, enum phy_mode mode,
>>   		}
>>   	}
>>   
>> +	port->role = submode;
>>   	mutex_unlock(&padctl->lock);
>>   
>>   	return err;
>> diff --git a/drivers/phy/tegra/xusb.h b/drivers/phy/tegra/xusb.h
>> index d2b5f9565132..273af147dfd3 100644
>> --- a/drivers/phy/tegra/xusb.h
>> +++ b/drivers/phy/tegra/xusb.h
>> @@ -317,6 +317,7 @@ struct tegra_xusb_usb2_port {
>>   	enum usb_dr_mode mode;
>>   	bool internal;
>>   	int usb3_port_fake;
>> +	enum usb_role role;
>>   };
> 
> Jonathan can we get some t-b for these two patches
>