Re: [PATCH V15 6/9] mfd: pm8008: Use i2c_new_dummy_device() API

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On 7/1/2022 1:24 PM, Lee Jones wrote:
On Fri, 01 Jul 2022, Satya Priya Kakitapalli (Temp) wrote:

On 6/30/2022 4:04 PM, Lee Jones wrote:
On Thu, 30 Jun 2022, Satya Priya Kakitapalli (Temp) wrote:

On 6/29/2022 8:48 PM, Lee Jones wrote:
On Wed, 29 Jun 2022, Satya Priya Kakitapalli (Temp) wrote:

On 6/28/2022 1:12 PM, Lee Jones wrote:
On Tue, 28 Jun 2022, Satya Priya Kakitapalli (Temp) wrote:

On 6/27/2022 1:11 PM, Lee Jones wrote:
On Mon, 27 Jun 2022, Satya Priya Kakitapalli (Temp) wrote:

Hi Lee,


On 6/20/2022 4:37 PM, Satya Priya Kakitapalli (Temp) wrote:
On 6/20/2022 1:50 PM, Lee Jones wrote:
On Mon, 20 Jun 2022, Satya Priya Kakitapalli (Temp) wrote:

On 6/17/2022 2:27 AM, Lee Jones wrote:
On Tue, 14 Jun 2022, Satya Priya wrote:

Use i2c_new_dummy_device() to register pm8008-regulator
client present at a different address space, instead of
defining a separate DT node. This avoids calling the probe
twice for the same chip, once for each client pm8008-infra
and pm8008-regulator.

As a part of this define pm8008_regmap_init() to do regmap
init for both the clients and define pm8008_get_regmap() to
pass the regmap to the regulator driver.

Signed-off-by: Satya Priya <quic_c_skakit@xxxxxxxxxxx>
Reviewed-by: Stephen Boyd <swboyd@xxxxxxxxxxxx>
---
Changes in V15:
        - None.

Changes in V14:
        - None.

Changes in V13:
        - None.

        drivers/mfd/qcom-pm8008.c       | 34
++++++++++++++++++++++++++++++++--
        include/linux/mfd/qcom_pm8008.h |  9 +++++++++
        2 files changed, 41 insertions(+), 2 deletions(-)
        create mode 100644 include/linux/mfd/qcom_pm8008.h

diff --git a/drivers/mfd/qcom-pm8008.c b/drivers/mfd/qcom-pm8008.c
index 569ffd50..55e2a8e 100644
--- a/drivers/mfd/qcom-pm8008.c
+++ b/drivers/mfd/qcom-pm8008.c
@@ -9,6 +9,7 @@
        #include <linux/interrupt.h>
        #include <linux/irq.h>
        #include <linux/irqdomain.h>
+#include <linux/mfd/qcom_pm8008.h>
        #include <linux/module.h>
        #include <linux/of_device.h>
        #include <linux/of_platform.h>
@@ -57,6 +58,7 @@ enum {
        struct pm8008_data {
            struct device *dev;
+    struct regmap *regulators_regmap;
            int irq;
            struct regmap_irq_chip_data *irq_data;
        };
@@ -150,6 +152,12 @@ static struct regmap_config
qcom_mfd_regmap_cfg = {
            .max_register    = 0xFFFF,
        };
+struct regmap *pm8008_get_regmap(const struct pm8008_data *chip)
+{
+    return chip->regulators_regmap;
+}
+EXPORT_SYMBOL_GPL(pm8008_get_regmap);
Seems like abstraction for the sake of abstraction.

Why not do the dereference inside the regulator driver?
To derefer this in the regulator driver, we need to have the
pm8008_data
struct definition in the qcom_pm8008 header file.

I think it doesn't look great to have only that structure in
header and all
other structs and enum in the mfd driver.
Then why pass 'pm8008_data' at all?
There is one more option, instead of passing the pm8008_data, we could
pass the pdev->dev.parent and get the pm8008 chip data directly in the
pm8008_get_regmap() like below


struct regmap *pm8008_get_regmap(const struct device *dev)
      {
          const struct pm8008_data *chip = dev_get_drvdata(dev);

          return chip->regulators_regmap;
}
EXPORT_SYMBOL_GPL(pm8008_get_regmap);


By doing this we can avoid having declaration of pm8008_data also in the
header. Please let me know if this looks good.

Could you please confirm on this?

What's preventing you from passing 'regmap'?
I didn't get what you meant here, could you please elaborate a bit?
Ah yes.  I authored you a patch, but became distracted. Here:

-----8<--------------------8<-------

From: Lee Jones <lee.jones@xxxxxxxxxx>

mfd: pm8008: Remove driver data structure pm8008_data
Maintaining a local driver data structure that is never shared
outside of the core device is an unnecessary complexity.  Half of the
attributes were not used outside of a single function, one of which
was not used at all.  The remaining 2 are generic and can be passed
around as required.
Okay, but we still need to store the regulators_regmap, which is required in
the pm8008 regulator driver. Could we use a global variable for it?
Look down ...

Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>
---
      drivers/mfd/qcom-pm8008.c | 53 ++++++++++++++++++-----------------------------
      1 file changed, 20 insertions(+), 33 deletions(-)

diff --git a/drivers/mfd/qcom-pm8008.c b/drivers/mfd/qcom-pm8008.c
index c472d7f8103c4..4b8ff947762f2 100644
--- a/drivers/mfd/qcom-pm8008.c
+++ b/drivers/mfd/qcom-pm8008.c
@@ -54,13 +54,6 @@ enum {
      #define PM8008_PERIPH_OFFSET(paddr)	(paddr - PM8008_PERIPH_0_BASE)
-struct pm8008_data {
-	struct device *dev;
-	struct regmap *regmap;
-	int irq;
-	struct regmap_irq_chip_data *irq_data;
-};
-
      static unsigned int p0_offs[] = {PM8008_PERIPH_OFFSET(PM8008_PERIPH_0_BASE)};
      static unsigned int p1_offs[] = {PM8008_PERIPH_OFFSET(PM8008_PERIPH_1_BASE)};
      static unsigned int p2_offs[] = {PM8008_PERIPH_OFFSET(PM8008_PERIPH_2_BASE)};
@@ -150,7 +143,7 @@ static struct regmap_config qcom_mfd_regmap_cfg = {
      	.max_register	= 0xFFFF,
      };
-static int pm8008_init(struct pm8008_data *chip)
+static int pm8008_init(struct regmap *regmap)
      {
      	int rc;
@@ -160,34 +153,31 @@ static int pm8008_init(struct pm8008_data *chip)
      	 * This is required to enable the writing of TYPE registers in
      	 * regmap_irq_sync_unlock().
      	 */
-	rc = regmap_write(chip->regmap,
-			 (PM8008_TEMP_ALARM_ADDR | INT_SET_TYPE_OFFSET),
-			 BIT(0));
+	rc = regmap_write(regmap, (PM8008_TEMP_ALARM_ADDR | INT_SET_TYPE_OFFSET), BIT(0));
      	if (rc)
      		return rc;
      	/* Do the same for GPIO1 and GPIO2 peripherals */
-	rc = regmap_write(chip->regmap,
-			 (PM8008_GPIO1_ADDR | INT_SET_TYPE_OFFSET), BIT(0));
+	rc = regmap_write(regmap, (PM8008_GPIO1_ADDR | INT_SET_TYPE_OFFSET), BIT(0));
      	if (rc)
      		return rc;
-	rc = regmap_write(chip->regmap,
-			 (PM8008_GPIO2_ADDR | INT_SET_TYPE_OFFSET), BIT(0));
+	rc = regmap_write(regmap, (PM8008_GPIO2_ADDR | INT_SET_TYPE_OFFSET), BIT(0));
      	return rc;
      }
-static int pm8008_probe_irq_peripherals(struct pm8008_data *chip,
+static int pm8008_probe_irq_peripherals(struct device *dev,
+					struct regmap *regmap,
      					int client_irq)
      {
      	int rc, i;
      	struct regmap_irq_type *type;
      	struct regmap_irq_chip_data *irq_data;
-	rc = pm8008_init(chip);
+	rc = pm8008_init(regmap);
      	if (rc) {
-		dev_err(chip->dev, "Init failed: %d\n", rc);
+		dev_err(dev, "Init failed: %d\n", rc);
      		return rc;
      	}
@@ -207,10 +197,10 @@ static int pm8008_probe_irq_peripherals(struct pm8008_data *chip,
      				IRQ_TYPE_LEVEL_HIGH | IRQ_TYPE_LEVEL_LOW);
      	}
-	rc = devm_regmap_add_irq_chip(chip->dev, chip->regmap, client_irq,
+	rc = devm_regmap_add_irq_chip(dev, regmap, client_irq,
      			IRQF_SHARED, 0, &pm8008_irq_chip, &irq_data);
      	if (rc) {
-		dev_err(chip->dev, "Failed to add IRQ chip: %d\n", rc);
+		dev_err(dev, "Failed to add IRQ chip: %d\n", rc);
      		return rc;
      	}
@@ -220,26 +210,23 @@ static int pm8008_probe_irq_peripherals(struct pm8008_data *chip,
      static int pm8008_probe(struct i2c_client *client)
      {
      	int rc;
-	struct pm8008_data *chip;
-
-	chip = devm_kzalloc(&client->dev, sizeof(*chip), GFP_KERNEL);
-	if (!chip)
-		return -ENOMEM;
+	struct device *dev;
+	struct regmap *regmap;
-	chip->dev = &client->dev;
-	chip->regmap = devm_regmap_init_i2c(client, &qcom_mfd_regmap_cfg);
-	if (!chip->regmap)
+	dev = &client->dev;
+	regmap = devm_regmap_init_i2c(client, &qcom_mfd_regmap_cfg);
+	if (!regmap)
      		return -ENODEV;
-	i2c_set_clientdata(client, chip);
+	i2c_set_clientdata(client, regmap);
Here ^
I have added a dummy device and set the client data by passing regmap, see
below:

+       regulators_client = i2c_new_dummy_device(client->adapter,
client->addr + 1);
+       if (IS_ERR(regulators_client)) {
+               dev_err(dev, "can't attach client\n");
+               return PTR_ERR(regulators_client);
+       }
+
+       regulators_regmap = devm_regmap_init_i2c(regulators_client,
&qcom_mfd_regmap_cfg[1]);
+       if (!regmap)
+               return -ENODEV;
+
+       i2c_set_clientdata(client, regulators_regmap);

Now if i try to get this regmap from regulator driver by doing

struct regmap *regmap = dev_get_drvdata(pdev->dev.parent);

it still gets me the regmap of pm8008@8 device and not the regulator device
regmap (0x9). Not sure if I'm missing something here.
So you need to pass 2 regmap pointers?

If you need to pass more than one item to the child devices, you do
need to use a struct for that.
I need to pass only one regmap out of the two, but i am not able to retrieve
the correct regmap simply by doing i2c_set_clientdata

probably because we are having all the child nodes under same DT node and
thus not able to distinguish based on the dev pointer
You can only pull out (get) the pointer that you put in (set).

Unless you over-wrote it later in the thread of execution, you are
pulling out whatever regulators_regmap happens to be.

Is qcom_mfd_regmap_cfg[1] definitely the one you want?

Yes, I need qcom_mfd_regmap_cfg[1]

Pasting code snippet for reference:

static struct regmap_config qcom_mfd_regmap_cfg[2] = {
      {

          .name = "infra",
          .reg_bits   = 16,
          .val_bits   = 8,
          .max_register   = 0xFFFF,
      },
      {
          .name = "regulators",
          .reg_bits   = 16,
          .val_bits   = 8,
          .max_register   = 0xFFFF,
      },

};


Inside pm8008_probe:


      regmap = devm_regmap_init_i2c(client, &qcom_mfd_regmap_cfg[0]);
      if (!regmap)
          return -ENODEV;

      i2c_set_clientdata(client, regmap);


      regulators_client = i2c_new_dummy_device(client->adapter, client->addr
+ 1);
      if (IS_ERR(regulators_client)) {
          dev_err(dev, "can't attach client\n");
          return PTR_ERR(regulators_client);
      }

      regulators_regmap = devm_regmap_init_i2c(regulators_client,
&qcom_mfd_regmap_cfg[1]);
      if (!regmap)
          return -ENODEV;

      i2c_set_clientdata(regulators_client, regulators_regmap);
You can't call this twice.

Doing so with over-write regmap with regulators_regmap.

You said you only needed one?

   "I need to pass only one regmap out of the two, but i am not able to retrieve"


I thought you asked whether we have to pass two regmaps to the child regulator driver.


In qcom-pm8008-regulator.c I tried to get the regmap using

dev_get_regmap(pdev->dev.parent, "regulators");
I haven't looked at this API before.  I suggest that this would be
used *instead* of passing the regmap pointer via driver_data.

It looks like you're using different devices to init your regmaps;
'client' and 'regulator_client' (derived from client->adapter).

"regulators" is registered using regulators_regmap which was *not*
init'ed with pdev->dev.parent (same as client->dev), so trying to
dev_get_regmap() with that device pointer will not work.


Okay, So I will leave the driver as is then?




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [Linux for Sparc]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux