[PATCH] pinctrl-sunxi: fix pin attribute handling.

[PATCH] pinctrl-sunxi: fix pin attribute handling.

[Ithamar R. Adema]

From: "Ithamar R. Adema" <ithamar@upgrade-android.com>

The configuration of pull up/down on pins does not allow disabling
of the pins. PIN_CONFIG_BIAS_PULL_UP and PIN_CONFIG_BIAS_PULL_DOWN
take an argument of 1 or 0 to specify enabling or disabling of the
pull up/down, but the code does not take this into account.

Also, default the pullup/downs to disabled if not specified, so no old
state from e.g. the bootloader is still active after reconfiguration
by the kernel.

Signed-off-by: Ithamar R. Adema <ithamar@upgrade-android.com>
---
 drivers/pinctrl/pinctrl-sunxi.c |   39 +++++++++++++++++----------------------
 1 file changed, 17 insertions(+), 22 deletions(-)

diff --git a/drivers/pinctrl/pinctrl-sunxi.c b/drivers/pinctrl/pinctrl-sunxi.c
index b7d8c89..a652061 100644
--- a/drivers/pinctrl/pinctrl-sunxi.c
+++ b/drivers/pinctrl/pinctrl-sunxi.c
@@ -1466,7 +1466,7 @@ static int sunxi_pctrl_dt_node_to_map(struct pinctrl_dev *pctldev,
 	of_property_for_each_string(node, "allwinner,pins", prop, group) {
 		struct sunxi_pinctrl_group *grp =
 			sunxi_pinctrl_find_group_by_name(pctl, group);
-		int j = 0, configlen = 0;
+		u16 strength;
 
 		if (!grp) {
 			dev_err(pctl->dev, "unknown pin %s", group);
@@ -1490,31 +1490,23 @@ static int sunxi_pctrl_dt_node_to_map(struct pinctrl_dev *pctldev,
 		(*map)[i].type = PIN_MAP_TYPE_CONFIGS_GROUP;
 		(*map)[i].data.configs.group_or_pin = group;
 
-		if (of_find_property(node, "allwinner,drive", NULL))
-			configlen++;
-		if (of_find_property(node, "allwinner,pull", NULL))
-			configlen++;
-
-		pinconfig = kzalloc(configlen * sizeof(*pinconfig), GFP_KERNEL);
+		pinconfig = kzalloc(3 * sizeof(*pinconfig), GFP_KERNEL);
 
+		strength = 0;
 		if (!of_property_read_u32(node, "allwinner,drive", &val)) {
-			u16 strength = (val + 1) * 10;
-			pinconfig[j++] =
-				pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
-							 strength);
+			strength = (val + 1) * 10;
 		}
+		pinconfig[0] =
+			pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
+						 strength);
 
-		if (!of_property_read_u32(node, "allwinner,pull", &val)) {
-			enum pin_config_param pull = PIN_CONFIG_END;
-			if (val == 1)
-				pull = PIN_CONFIG_BIAS_PULL_UP;
-			else if (val == 2)
-				pull = PIN_CONFIG_BIAS_PULL_DOWN;
-			pinconfig[j++] = pinconf_to_config_packed(pull, 0);
-		}
+		val = 0;
+		of_property_read_u32(node, "allwinner,pull", &val);
+		pinconfig[1] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_UP, val == 1);
+		pinconfig[2] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_DOWN, val == 2);
 
 		(*map)[i].data.configs.configs = pinconfig;
-		(*map)[i].data.configs.num_configs = configlen;
+		(*map)[i].data.configs.num_configs = 3;
 
 		i++;
 	}
@@ -1563,6 +1555,7 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
 {
 	struct sunxi_pinctrl *pctl = pinctrl_dev_get_drvdata(pctldev);
 	struct sunxi_pinctrl_group *g = &pctl->groups[group];
+	u32 arg = pinconf_to_config_argument(config);
 	u32 val, mask;
 	u16 strength;
 	u8 dlevel;
@@ -1586,15 +1579,17 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
 			pctl->membase + sunxi_dlevel_reg(g->pin));
 		break;
 	case PIN_CONFIG_BIAS_PULL_UP:
+		arg = !!arg;
 		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
 		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
-		writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin),
+		writel((val & ~mask) | arg << sunxi_pull_offset(g->pin),
 			pctl->membase + sunxi_pull_reg(g->pin));
 		break;
 	case PIN_CONFIG_BIAS_PULL_DOWN:
+		arg = !!arg;
 		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
 		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
-		writel((val & ~mask) | 2 << sunxi_pull_offset(g->pin),
+		writel((val & ~mask) | (arg << 1) << sunxi_pull_offset(g->pin),
 			pctl->membase + sunxi_pull_reg(g->pin));
 		break;
 	default:
-- 
1.7.9.5

[PATCH] pinctrl-sunxi: fix pin attribute handling.

[Maxime Ripard]

Hi Ithamar,

You should probably put in the recipients Linus Walleij that will
probably be the one that will merge this patch anyway.

On Thu, Jun 20, 2013 at 07:12:53PM +0200, Ithamar R. Adema wrote:
> From: "Ithamar R. Adema" <ithamar@upgrade-android.com>
> 
> The configuration of pull up/down on pins does not allow disabling
> of the pins. PIN_CONFIG_BIAS_PULL_UP and PIN_CONFIG_BIAS_PULL_DOWN
> take an argument of 1 or 0 to specify enabling or disabling of the
> pull up/down, but the code does not take this into account.
> 
> Also, default the pullup/downs to disabled if not specified, so no old
> state from e.g. the bootloader is still active after reconfiguration
> by the kernel.
> 
> Signed-off-by: Ithamar R. Adema <ithamar@upgrade-android.com>
> ---
>  drivers/pinctrl/pinctrl-sunxi.c |   39 +++++++++++++++++----------------------
>  1 file changed, 17 insertions(+), 22 deletions(-)
> 
> diff --git a/drivers/pinctrl/pinctrl-sunxi.c b/drivers/pinctrl/pinctrl-sunxi.c
> index b7d8c89..a652061 100644
> --- a/drivers/pinctrl/pinctrl-sunxi.c
> +++ b/drivers/pinctrl/pinctrl-sunxi.c
> @@ -1466,7 +1466,7 @@ static int sunxi_pctrl_dt_node_to_map(struct pinctrl_dev *pctldev,
>  	of_property_for_each_string(node, "allwinner,pins", prop, group) {
>  		struct sunxi_pinctrl_group *grp =
>  			sunxi_pinctrl_find_group_by_name(pctl, group);
> -		int j = 0, configlen = 0;
> +		u16 strength;
>  
>  		if (!grp) {
>  			dev_err(pctl->dev, "unknown pin %s", group);
> @@ -1490,31 +1490,23 @@ static int sunxi_pctrl_dt_node_to_map(struct pinctrl_dev *pctldev,
>  		(*map)[i].type = PIN_MAP_TYPE_CONFIGS_GROUP;
>  		(*map)[i].data.configs.group_or_pin = group;
>  
> -		if (of_find_property(node, "allwinner,drive", NULL))
> -			configlen++;
> -		if (of_find_property(node, "allwinner,pull", NULL))
> -			configlen++;
> -
> -		pinconfig = kzalloc(configlen * sizeof(*pinconfig), GFP_KERNEL);
> +		pinconfig = kzalloc(3 * sizeof(*pinconfig), GFP_KERNEL);
>  
> +		strength = 0;
>  		if (!of_property_read_u32(node, "allwinner,drive", &val)) {
> -			u16 strength = (val + 1) * 10;
> -			pinconfig[j++] =
> -				pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
> -							 strength);
> +			strength = (val + 1) * 10;
>  		}
> +		pinconfig[0] =
> +			pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
> +						 strength);

Why are you making the configuration of the strength in any cases? It's
an optional property, it should be treated as such.

> -		if (!of_property_read_u32(node, "allwinner,pull", &val)) {
> -			enum pin_config_param pull = PIN_CONFIG_END;
> -			if (val == 1)
> -				pull = PIN_CONFIG_BIAS_PULL_UP;
> -			else if (val == 2)
> -				pull = PIN_CONFIG_BIAS_PULL_DOWN;
> -			pinconfig[j++] = pinconf_to_config_packed(pull, 0);
> -		}
> +		val = 0;
> +		of_property_read_u32(node, "allwinner,pull", &val);
> +		pinconfig[1] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_UP, val == 1);
> +		pinconfig[2] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_DOWN, val == 2);
 
I'm not sure this is actually right. The pins are either configured to
have a pull-up, a pull-down, or none of it, in the same register, in the
same bitfield, so why adding it in any cases? we only care about at most
one of the two, and those two options are mutually exclusive.

>  		(*map)[i].data.configs.configs = pinconfig;
> -		(*map)[i].data.configs.num_configs = configlen;
> +		(*map)[i].data.configs.num_configs = 3;
>  
>  		i++;
>  	}
> @@ -1563,6 +1555,7 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
>  {
>  	struct sunxi_pinctrl *pctl = pinctrl_dev_get_drvdata(pctldev);
>  	struct sunxi_pinctrl_group *g = &pctl->groups[group];
> +	u32 arg = pinconf_to_config_argument(config);
>  	u32 val, mask;
>  	u16 strength;
>  	u8 dlevel;
> @@ -1586,15 +1579,17 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
>  			pctl->membase + sunxi_dlevel_reg(g->pin));
>  		break;
>  	case PIN_CONFIG_BIAS_PULL_UP:
> +		arg = !!arg;

Ouch, nope.

Put this next to the call to pinconf_to_config_argument

>  		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
>  		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
> -		writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin),
> +		writel((val & ~mask) | arg << sunxi_pull_offset(g->pin),

I'd rather see an obvious assignment of 1 here. It's the value that is
documented in the datasheet, it pops to the mind. You can always put
this inside an if(arg) statement if you want to.

>  			pctl->membase + sunxi_pull_reg(g->pin));
>  		break;
>  	case PIN_CONFIG_BIAS_PULL_DOWN:
> +		arg = !!arg;
>  		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
>  		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
> -		writel((val & ~mask) | 2 << sunxi_pull_offset(g->pin),
> +		writel((val & ~mask) | (arg << 1) << sunxi_pull_offset(g->pin),

Ditto.

>  			pctl->membase + sunxi_pull_reg(g->pin));
>  		break;
>  	default:

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[PATCH] pinctrl-sunxi: fix pin attribute handling.

[Ithamar R. Adema]

Dear Maxime,

On Jun 20, 2013, at 10:49 , Maxime Ripard <maxime.ripard@free-electrons.com> wrote:

> Hi Ithamar,
> 
> You should probably put in the recipients Linus Walleij that will
> probably be the one that will merge this patch anyway.

Ah okay, thanks for the heads-up.
>> +		strength = 0;
>> 		if (!of_property_read_u32(node, "allwinner,drive", &val)) {
>> -			u16 strength = (val + 1) * 10;
>> -			pinconfig[j++] =
>> -				pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
>> -							 strength);
>> +			strength = (val + 1) * 10;
>> 		}
>> +		pinconfig[0] =
>> +			pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
>> +						 strength);
> 
> Why are you making the configuration of the strength in any cases? It's
> an optional property, it should be treated as such.

My reasoning, which I can see is very subjective, is that if those properties (same for pull up/down) are not specified, you still want them to be in a defined state. Setting the drive strength (and pull up/down) to the default 0 value is just so that any configuration that might have been done the kernel started is undone.

I can understand your reasoning here, but are you sure you would want them left in undefined state?

> 
>> -		if (!of_property_read_u32(node, "allwinner,pull", &val)) {
>> -			enum pin_config_param pull = PIN_CONFIG_END;
>> -			if (val == 1)
>> -				pull = PIN_CONFIG_BIAS_PULL_UP;
>> -			else if (val == 2)
>> -				pull = PIN_CONFIG_BIAS_PULL_DOWN;
>> -			pinconfig[j++] = pinconf_to_config_packed(pull, 0);
>> -		}
>> +		val = 0;
>> +		of_property_read_u32(node, "allwinner,pull", &val);
>> +		pinconfig[1] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_UP, val == 1);
>> +		pinconfig[2] = pinconf_to_config_packed(PIN_CONFIG_BIAS_PULL_DOWN, val == 2);
> 
> I'm not sure this is actually right. The pins are either configured to
> have a pull-up, a pull-down, or none of it, in the same register, in the
> same bitfield, so why adding it in any cases? we only care about at most
> one of the two, and those two options are mutually exclusive.

True, this is my fault. I guess I thought that settings the pull up didn't clear the pull down, but after looking at the code, this is properly handled. I'll fix this properly in the next version of the patch.

> 
>> 		(*map)[i].data.configs.configs = pinconfig;
>> -		(*map)[i].data.configs.num_configs = configlen;
>> +		(*map)[i].data.configs.num_configs = 3;
>> 
>> 		i++;
>> 	}
>> @@ -1563,6 +1555,7 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
>> {
>> 	struct sunxi_pinctrl *pctl = pinctrl_dev_get_drvdata(pctldev);
>> 	struct sunxi_pinctrl_group *g = &pctl->groups[group];
>> +	u32 arg = pinconf_to_config_argument(config);
>> 	u32 val, mask;
>> 	u16 strength;
>> 	u8 dlevel;
>> @@ -1586,15 +1579,17 @@ static int sunxi_pconf_group_set(struct pinctrl_dev *pctldev,
>> 			pctl->membase + sunxi_dlevel_reg(g->pin));
>> 		break;
>> 	case PIN_CONFIG_BIAS_PULL_UP:
>> +		arg = !!arg;
> 
> Ouch, nope.
> 
> Put this next to the call to pinconf_to_config_argument

Fair enough, will fix.
> 
>> 		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
>> 		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
>> -		writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin),
>> +		writel((val & ~mask) | arg << sunxi_pull_offset(g->pin),
> 
> I'd rather see an obvious assignment of 1 here. It's the value that is
> documented in the datasheet, it pops to the mind. You can always put
> this inside an if(arg) statement if you want to.

Just to make sure I understand, are you suggesting to stretch that writel() onliner into an if (arg) writel() else writel() ?

Thanks for taking the time to review!

Regards,

Ithamar.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4163 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130621/1f7ae23f/attachment-0001.p7s>

[PATCH] pinctrl-sunxi: fix pin attribute handling.

[Maxime Ripard]

Hi Ithamar,

On Fri, Jun 21, 2013 at 09:41:13AM +0200, Ithamar R. Adema wrote:
> Dear Maxime,
> 
> On Jun 20, 2013, at 10:49 , Maxime Ripard <maxime.ripard@free-electrons.com> wrote:
> 
> > Hi Ithamar,
> > 
> > You should probably put in the recipients Linus Walleij that will
> > probably be the one that will merge this patch anyway.
> 
> Ah okay, thanks for the heads-up.
> >> +		strength = 0;
> >> 		if (!of_property_read_u32(node, "allwinner,drive", &val)) {
> >> -			u16 strength = (val + 1) * 10;
> >> -			pinconfig[j++] =
> >> -				pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
> >> -							 strength);
> >> +			strength = (val + 1) * 10;
> >> 		}
> >> +		pinconfig[0] =
> >> +			pinconf_to_config_packed(PIN_CONFIG_DRIVE_STRENGTH,
> >> +						 strength);
> > 
> > Why are you making the configuration of the strength in any cases? It's
> > an optional property, it should be treated as such.
> 
> My reasoning, which I can see is very subjective, is that if those
> properties (same for pull up/down) are not specified, you still want
> them to be in a defined state. Setting the drive strength (and pull
> up/down) to the default 0 value is just so that any configuration that
> might have been done the kernel started is undone.
> 
> I can understand your reasoning here, but are you sure you would want
> them left in undefined state?

While your reasoning is correct here, the point is that the platform is
still a little obscure, with not a lot of documentation and so on. So
relying on the bootloader to do some muxing could still prove useful.

> > 
> >> 		val = readl(pctl->membase + sunxi_pull_reg(g->pin));
> >> 		mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
> >> -		writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin),
> >> +		writel((val & ~mask) | arg << sunxi_pull_offset(g->pin),
> > 
> > I'd rather see an obvious assignment of 1 here. It's the value that is
> > documented in the datasheet, it pops to the mind. You can always put
> > this inside an if(arg) statement if you want to.
> 
> Just to make sure I understand, are you suggesting to stretch that
> writel() onliner into an if (arg) writel() else writel() ?

Yes, something like

case PULL_UP:
    val = readl(pctl->membase + sunxi_pull_reg(g->pin));
    mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
    if (arg)
	writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin), reg);
    else
	writel(val & ~mask, reg);

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130623/02b42f03/attachment.sig>

[PATCH] pinctrl-sunxi: fix pin attribute handling.

[Ithamar R. Adema]

Dear Maxime,

On Jun 23, 2013, at 12:20 , Maxime Ripard <maxime.ripard@free-electrons.com> wrote:

> While your reasoning is correct here, the point is that the platform is
> still a little obscure, with not a lot of documentation and so on. So
> relying on the bootloader to do some muxing could still prove useful.

Fair enough, I'll adjust this for sunxi then. The reason I'm asking is because I used this code for another architecture I hope to mainline (telechips) and I just wanted to know if this was some requirement of the subsystem or more related to the platform. Thanks for your explanation.

>> Just to make sure I understand, are you suggesting to stretch that
>> writel() onliner into an if (arg) writel() else writel() ?
> 
> Yes, something like
> 
> case PULL_UP:
>    val = readl(pctl->membase + sunxi_pull_reg(g->pin));
>    mask = PULL_PINS_MASK << sunxi_pull_offset(g->pin);
>    if (arg)
> 	writel((val & ~mask) | 1 << sunxi_pull_offset(g->pin), reg);
>    else
> 	writel(val & ~mask, reg);

Okay, clear.

I'll send a new patch along fixes the mentioned issues as suggested.

Thanks!

Regards,

Ithamar.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4163 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130623/91f87caa/attachment.p7s>