[RFC PATCH] usb: dwc2: Try usb_get_phy_by_phandle instead of usb_get_phy

[RFC PATCH] usb: dwc2: Try usb_get_phy_by_phandle instead of usb_get_phy

[Jules Maselbas]

On probe the dwc2 driver tries two path to get an usb phy, first calling
devm_phy_get() and secondly with devm_usb_get_phy().

However the current implementation of devm_phy_get() never return a valid
phy for usb-nop-xceiv. And the current implementation of devm_usb_get_phy
only check for phy that's has already been registered.

During boot, I see the dwc2 driver being probed before the usb-nop-xceiv
probe, this means that during the dwc2 probe the function devm_usb_get_phy
never finds the a phy (because it hasn't been registered yet) but never
cause the dwc2 probe to defer.

I tried to follow what is done by dwc3_core_get_phy(): if the current
device has an of_node then try to get the usb_phy by phandle instead of
using devm_usb_get_phy(). This way when the probe order is not good the
devm_usb_get_phy_by_phandle() function will return -EPROBE_DEFER.

Signed-off-by: Jules Maselbas <jmaselbas@kalray.eu>
--- 8< ---

A snippet of the device-tree source I am using:
        &usb0 {
                phys = <&usb_phy0>;
                phy-names = "usb2-phy";
        };
        &usb_phy0 {
                #phy-cells = <0>;
                compatible = "usb-nop-xceiv";
                reset-gpios = <&gpio 0 GPIO_ACTIVE_LOW>;
        };
---
 drivers/usb/dwc2/platform.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/dwc2/platform.c b/drivers/usb/dwc2/platform.c
index db9fd4bd1a38..b58ce996add7 100644
--- a/drivers/usb/dwc2/platform.c
+++ b/drivers/usb/dwc2/platform.c
@@ -251,7 +251,12 @@ static int dwc2_lowlevel_hw_init(struct dwc2_hsotg *hsotg)
 	}
 
 	if (!hsotg->phy) {
-		hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
+		if (hsotg->dev->of_node)
+			i = of_property_match_string(hsotg->dev->of_node, "phy-names", "usb2-phy");
+		if (hsotg->dev->of_node && i >= 0)
+			hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", i);
+		else
+			hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
 		if (IS_ERR(hsotg->uphy)) {
 			ret = PTR_ERR(hsotg->uphy);
 			switch (ret) {
-- 
2.17.1

Re: [RFC PATCH] usb: dwc2: Try usb_get_phy_by_phandle instead of usb_get_phy

[Artur Petrosyan]

Hi Jules,

On 12/16/2020 20:59, Jules Maselbas wrote:
> On probe the dwc2 driver tries two path to get an usb phy, first calling
> devm_phy_get() and secondly with devm_usb_get_phy().
> 
> However the current implementation of devm_phy_get() never return a valid
> phy for usb-nop-xceiv. And the current implementation of devm_usb_get_phy
> only check for phy that's has already been registered.
> 
> During boot, I see the dwc2 driver being probed before the usb-nop-xceiv
> probe, this means that during the dwc2 probe the function devm_usb_get_phy
> never finds the a phy (because it hasn't been registered yet) but never
> cause the dwc2 probe to defer.
> 
> I tried to follow what is done by dwc3_core_get_phy(): if the current
> device has an of_node then try to get the usb_phy by phandle instead of
> using devm_usb_get_phy(). This way when the probe order is not good the
> devm_usb_get_phy_by_phandle() function will return -EPROBE_DEFER.
> 
> Signed-off-by: Jules Maselbas <jmaselbas@kalray.eu>
> --- 8< ---
> 
> A snippet of the device-tree source I am using:
>          &usb0 {
>                  phys = <&usb_phy0>;
>                  phy-names = "usb2-phy";
>          };
>          &usb_phy0 {
>                  #phy-cells = <0>;
>                  compatible = "usb-nop-xceiv";
>                  reset-gpios = <&gpio 0 GPIO_ACTIVE_LOW>;
>          };
> ---
>   drivers/usb/dwc2/platform.c | 7 ++++++-
>   1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/usb/dwc2/platform.c b/drivers/usb/dwc2/platform.c
> index db9fd4bd1a38..b58ce996add7 100644
> --- a/drivers/usb/dwc2/platform.c
> +++ b/drivers/usb/dwc2/platform.c
> @@ -251,7 +251,12 @@ static int dwc2_lowlevel_hw_init(struct dwc2_hsotg *hsotg)
>   	}
>   
>   	if (!hsotg->phy) {
> -		hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
> +		if (hsotg->dev->of_node)
> +			i = of_property_match_string(hsotg->dev->of_node, "phy-names", "usb2-phy");

According the device tree you have provided the value of "i" will always 
be "0".

> +		if (hsotg->dev->of_node && i >= 0)
> +			hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", i);

Why do you use the value of "i" while in "<&usb_phy0>" you have only one 
phy. If you had several phy-names and the value of "i" gets more than 0, 
then based on your usb_phy0 "devm_usb_get_phy_by_phandle" function will 
return error. So, maybe it would be more correct (based on your device 
tree), to use below command
hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", 0);

> +		else
> +			hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
>   		if (IS_ERR(hsotg->uphy)) {
>   			ret = PTR_ERR(hsotg->uphy);
>   			switch (ret) {
> 

Regards,
Artur

Re: [RFC PATCH] usb: dwc2: Try usb_get_phy_by_phandle instead of usb_get_phy

[Jules Maselbas]

Hi Artur,

On Fri, Dec 25, 2020 at 11:41:04AM +0000, Artur Petrosyan wrote:
> > @@ -251,7 +251,12 @@ static int dwc2_lowlevel_hw_init(struct dwc2_hsotg *hsotg)
> >   	}
> >   
> >   	if (!hsotg->phy) {
> > -		hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
> > +		if (hsotg->dev->of_node)
> > +			i = of_property_match_string(hsotg->dev->of_node, "phy-names", "usb2-phy");
> 
> According the device tree you have provided the value of "i" will always 
> be "0".
Yes

> > +		if (hsotg->dev->of_node && i >= 0)
> > +			hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", i);
> 
> Why do you use the value of "i" while in "<&usb_phy0>" you have only one 
> phy. If you had several phy-names and the value of "i" gets more than 0, 
> then based on your usb_phy0 "devm_usb_get_phy_by_phandle" function will 
> return error. So, maybe it would be more correct (based on your device 
> tree), to use below command
> hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", 0);
Yes I could use 0 instead of i, but I would like this to work not only
for my case where the "usb2-phy" phandle comes first.

I've tried to follow what's done in phy-core.c, as done by the function
phy_get. Where it first call "of_property_match_string" and then get the
phy with the matched index.

I don't see how, in my case, the function "devm_usb_get_phy_by_phandle" can
be called with i greater than 0, and returning an error.

Best,
Jules

Re: [RFC PATCH] usb: dwc2: Try usb_get_phy_by_phandle instead of usb_get_phy

[Artur Petrosyan]

Hi Jules,

On 12/26/2020 17:45, Jules Maselbas wrote:
> Hi Artur,
> 
> On Fri, Dec 25, 2020 at 11:41:04AM +0000, Artur Petrosyan wrote:
>>> @@ -251,7 +251,12 @@ static int dwc2_lowlevel_hw_init(struct dwc2_hsotg *hsotg)
>>>    	}
>>>    
>>>    	if (!hsotg->phy) {
>>> -		hsotg->uphy = devm_usb_get_phy(hsotg->dev, USB_PHY_TYPE_USB2);
>>> +		if (hsotg->dev->of_node)
>>> +			i = of_property_match_string(hsotg->dev->of_node, "phy-names", "usb2-phy");
>>
>> According the device tree you have provided the value of "i" will always
>> be "0".
> Yes
> 
>>> +		if (hsotg->dev->of_node && i >= 0)
>>> +			hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", i);
>>
>> Why do you use the value of "i" while in "<&usb_phy0>" you have only one
>> phy. If you had several phy-names and the value of "i" gets more than 0,
>> then based on your usb_phy0 "devm_usb_get_phy_by_phandle" function will
>> return error. So, maybe it would be more correct (based on your device
>> tree), to use below command
>> hsotg->uphy = devm_usb_get_phy_by_phandle(hsotg->dev, "phys", 0);
> Yes I could use 0 instead of i, but I would like this to work not only
> for my case where the "usb2-phy" phandle comes first.
> 
> I've tried to follow what's done in phy-core.c, as done by the function
> phy_get. Where it first call "of_property_match_string" and then get the
> phy with the matched index.
> 
> I don't see how, in my case, the function "devm_usb_get_phy_by_phandle" can
> be called with i greater than 0, and returning an error.

I have noticed that there might be no need to check for "i >= 0" at all 
because "of_parse_phandle()" function has a check for index to not be 
less then 0.

> 
> Best,
> Jules
> 

Regards,
Artur