Re: [PATCH 1/3] hwmon: max31827: Add PEC support

Guenter Roeck <linux@xxxxxxxxxxxx> · Fri, 15 Dec 2023 17:32:49 -0800

On 12/15/23 12:28, Matyas, Daniel wrote:


-----Original Message-----
From: Guenter Roeck <groeck7@xxxxxxxxx> On Behalf Of Guenter Roeck
Sent: Thursday, December 14, 2023 6:10 PM
To: Matyas, Daniel <Daniel.Matyas@xxxxxxxxxx>
Cc: Jean Delvare <jdelvare@xxxxxxxx>; Rob Herring
<robh+dt@xxxxxxxxxx>; Krzysztof Kozlowski
<krzysztof.kozlowski+dt@xxxxxxxxxx>; Conor Dooley
<conor+dt@xxxxxxxxxx>; Jonathan Corbet <corbet@xxxxxxx>; linux-
hwmon@xxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-
kernel@xxxxxxxxxxxxxxx; linux-doc@xxxxxxxxxxxxxxx
Subject: Re: [PATCH 1/3] hwmon: max31827: Add PEC support

[External]

On 12/14/23 06:36, Daniel Matyas wrote:
Removed regmap and used my functions to read, write and update bits.
In these functions i2c_smbus_ helper functions are used. These check
if there were any PEC errors during read. In the write function, if
PEC is enabled, I check for PEC Error bit, to see if there were any errors.

Signed-off-by: Daniel Matyas <daniel.matyas@xxxxxxxxxx>

The "PEC" attribute needs to be attached to the I2C device.
See lm90.c or pmbus_core.c for examples.


I added pec_show() and pec_store() functions and created the pec file within the max31827_groups.
I did not set the flags, because I want them to be set only in pec_store. By default the PEC flag should not be set.


That is not the point. Again,

>> The "PEC" attribute needs to be attached to the I2C device.
>> See lm90.c or pmbus_core.c for examples.

That is not about regmap, it is about the location of the "pec" attribute.

The changes, if indeed needed, do not warant dropping regmap.
You will need to explain why the reg_read and reg_write callbacks
provideed by regmap can not be used.

On top of that, it is not clear why regmap can't be used in the first place.
It seems that the major change is that one needs to read the configuration
register after a write to see if there was a PEC error. It is not immediately
obvious why that additional read (if indeed necessary) would require
regmap support to be dropped.


So I am not sure about this, but it seems to me, that regmap_write is not sending a PEC byte and regmap_read is not checking the PEC byte by default. My rationale was: i2c_smbus_write_word_data() is using the i2c_xfer function, which has as argument the client->flags. So, if I2C_CLIENT_PEC is set in client->flags, that would be transmitted by i2c_xfer. I am not convinced about this, but lm90 and pmbus_core are not using regmap, so I went ahead and replaced it.

If what I am thinking is wrong, please correct me.


regmap uses smbus functions to access the chip if the underlying i2c driver
supports SMBus word operations. I fail to see the difference to your code.
Sure, max31827_reg_write() executes another read after the write, but that
is again a 16-bit operation and might we well be implemented as another
regmap operation to read the status register.

It would be possible to replace the regmap i2c code, use raw regmap
code instead, and provide ->read and ->write callbacks in struct regmap_config,
but I am not convinced that this would be beneficial.

Either case, sorry, I can not follow your line of argument.

Regarding "if indeed necessary": There needs to be a comment explaining
that the device will return ACK even after a packet with bad PEC is
received.

---
   Documentation/hwmon/max31827.rst |  14 +-
   drivers/hwmon/max31827.c         | 219 +++++++++++++++++++++++-----
---
   2 files changed, 171 insertions(+), 62 deletions(-)

diff --git a/Documentation/hwmon/max31827.rst
b/Documentation/hwmon/max31827.rst
index 44ab9dc064cb..ecbc1ddba6a7 100644
--- a/Documentation/hwmon/max31827.rst
+++ b/Documentation/hwmon/max31827.rst
@@ -131,7 +131,13 @@ The Fault Queue bits select how many
consecutive temperature faults must occur
   before overtemperature or undertemperature faults are indicated in
the
   corresponding status bits.

-Notes
------
-
-PEC is not implemented.
+PEC (packet error checking) can be enabled from the "pec" device
attribute.
+If PEC is enabled, a PEC byte is appended to the end of each message
transfer.
+This is a CRC-8 byte that is calculated on all of the message bytes
+(including the address/read/write byte). The last device to transmit
+a data byte also transmits the PEC byte. The master transmits the PEC
+byte after a write transaction, and the MAX31827 transmits the PEC
byte after a read transaction.
+
+The read PEC error is handled inside the
i2c_smbus_read_word_swapped() function.
+To check if the write had any PEC error a read is performed on the
+configuration register, to check the PEC Error bit.
\ No newline at end of file
diff --git a/drivers/hwmon/max31827.c b/drivers/hwmon/max31827.c
index
71ad3934dfb6..db93492193bd 100644
--- a/drivers/hwmon/max31827.c
+++ b/drivers/hwmon/max31827.c
@@ -11,8 +11,8 @@
   #include <linux/hwmon.h>
   #include <linux/i2c.h>
   #include <linux/mutex.h>
-#include <linux/regmap.h>
   #include <linux/regulator/consumer.h>
+#include <linux/hwmon-sysfs.h>
   #include <linux/of_device.h>

   #define MAX31827_T_REG			0x0
@@ -24,11 +24,13 @@

   #define MAX31827_CONFIGURATION_1SHOT_MASK	BIT(0)
   #define MAX31827_CONFIGURATION_CNV_RATE_MASK
	GENMASK(3, 1)
+#define MAX31827_CONFIGURATION_PEC_EN_MASK	BIT(4)
   #define MAX31827_CONFIGURATION_TIMEOUT_MASK	BIT(5)
   #define MAX31827_CONFIGURATION_RESOLUTION_MASK
	GENMASK(7, 6)
   #define MAX31827_CONFIGURATION_ALRM_POL_MASK	BIT(8)
   #define MAX31827_CONFIGURATION_COMP_INT_MASK	BIT(9)
   #define MAX31827_CONFIGURATION_FLT_Q_MASK	GENMASK(11, 10)
+#define MAX31827_CONFIGURATION_PEC_ERR_MASK	BIT(13)
   #define MAX31827_CONFIGURATION_U_TEMP_STAT_MASK	BIT(14)
   #define MAX31827_CONFIGURATION_O_TEMP_STAT_MASK	BIT(15)

@@ -94,23 +96,67 @@ struct max31827_state {
   	 * Prevent simultaneous access to the i2c client.
   	 */
   	struct mutex lock;
-	struct regmap *regmap;
   	bool enable;
   	unsigned int resolution;
   	unsigned int update_interval;
+	struct i2c_client *client;
   };

-static const struct regmap_config max31827_regmap = {
-	.reg_bits = 8,
-	.val_bits = 16,
-	.max_register = 0xA,
-};
+static int max31827_reg_read(struct i2c_client *client, u8 reg, u16
+*val) {
+	u16 tmp = i2c_smbus_read_word_swapped(client, reg);
+
+	if (tmp < 0)

An u16 variable will never be negative.

+		return tmp;
+
+	*val = tmp;
+	return 0;
+}

If regmap can indeed not be used, it is unnecessary to provide a pointer to
the return value. Instead, just like with smbus calls, the error return and
the return value can be combined. Adding this function just to separate
error from return value adds zero value (and, as can be seen from the
above, actually adds an opportunity to introduce bugs).

+
+static int max31827_reg_write(struct i2c_client *client, u8 reg, u16
+val) {
+	u16 cfg;
+	int ret;
+
+	ret = i2c_smbus_write_word_swapped(client, reg, val);
+	if (ret)
+		return ret;
+
+	// If PEC is not enabled, return with success

Do not mix comment styles. The rest of the driver doesn't use C++
comments.
Besides, the comment does not add any value.

+	if (!(client->flags & I2C_CLIENT_PEC))
+		return 0;
+
+	ret = max31827_reg_read(client,
MAX31827_CONFIGURATION_REG, &cfg);
+	if (ret)
+		return ret;
+
+	if (cfg & MAX31827_CONFIGURATION_PEC_ERR_MASK)
+		return -EBADMSG;
+

EBADMSG is "Not a data message". I don't think that is appropriate here.

I would very much prefer
	if (client->flags & I2C_CLIENT_PEC) {
		...
	}

+	return 0;
+}
+
+static int max31827_update_bits(struct i2c_client *client, u8 reg,
+				u16 mask, u16 val)
+{
+	u16 tmp;
+	int ret;
+
+	ret = max31827_reg_read(client, reg, &tmp);
+	if (ret)
+		return ret;
+
+	tmp = (tmp & ~mask) | (val & mask);
+	ret = max31827_reg_write(client, reg, tmp);
+
+	return ret;
+}

   static int shutdown_write(struct max31827_state *st, unsigned int reg,
   			  unsigned int mask, unsigned int val)
   {
-	unsigned int cfg;
-	unsigned int cnv_rate;
+	u16 cfg;
+	u16 cnv_rate;

I really do not see the point of those changes. u16 is more expensive than
unsigned int on many architectures. If retained, you'll have to explain why
this is needed and beneficial.

   	int ret;

   	/*
@@ -125,34 +171,34 @@ static int shutdown_write(struct
max31827_state
*st, unsigned int reg,

   	if (!st->enable) {
   		if (!mask)
-			ret = regmap_write(st->regmap, reg, val);
+			ret = max31827_reg_write(st->client, reg, val);
   		else
-			ret = regmap_update_bits(st->regmap, reg, mask,
val);
+			ret = max31827_update_bits(st->client, reg,
mask, val);
   		goto unlock;
   	}

-	ret = regmap_read(st->regmap,
MAX31827_CONFIGURATION_REG, &cfg);
+	ret = max31827_reg_read(st->client,
MAX31827_CONFIGURATION_REG,
+&cfg);
   	if (ret)
   		goto unlock;

   	cnv_rate = MAX31827_CONFIGURATION_CNV_RATE_MASK & cfg;
   	cfg = cfg & ~(MAX31827_CONFIGURATION_1SHOT_MASK |
   		      MAX31827_CONFIGURATION_CNV_RATE_MASK);
-	ret = regmap_write(st->regmap,
MAX31827_CONFIGURATION_REG, cfg);
+	ret = max31827_reg_write(st->client,
MAX31827_CONFIGURATION_REG,
+cfg);
   	if (ret)
   		goto unlock;

   	if (!mask)
-		ret = regmap_write(st->regmap, reg, val);
+		ret = max31827_reg_write(st->client, reg, val);
   	else
-		ret = regmap_update_bits(st->regmap, reg, mask, val);
+		ret = max31827_update_bits(st->client, reg, mask, val);

   	if (ret)
   		goto unlock;

-	ret = regmap_update_bits(st->regmap,
MAX31827_CONFIGURATION_REG,
-
MAX31827_CONFIGURATION_CNV_RATE_MASK,
-				 cnv_rate);
+	ret = max31827_update_bits(st->client,
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_CNV_RATE_MASK,
+				   cnv_rate);

   unlock:
   	mutex_unlock(&st->lock);
@@ -198,15 +244,16 @@ static int max31827_read(struct device *dev,
enum hwmon_sensor_types type,
   			 u32 attr, int channel, long *val)
   {
   	struct max31827_state *st = dev_get_drvdata(dev);
-	unsigned int uval;
+	u16 uval;
   	int ret = 0;

   	switch (type) {
   	case hwmon_temp:
   		switch (attr) {
   		case hwmon_temp_enable:
-			ret = regmap_read(st->regmap,
-
MAX31827_CONFIGURATION_REG, &uval);
+			ret = max31827_reg_read(st->client,
+
	MAX31827_CONFIGURATION_REG,
+						&uval);
   			if (ret)
   				break;

@@ -226,10 +273,10 @@ static int max31827_read(struct device *dev,
enum hwmon_sensor_types type,
   				 * be changed during the conversion
process.
   				 */

-				ret = regmap_update_bits(st->regmap,
-
MAX31827_CONFIGURATION_REG,
-
MAX31827_CONFIGURATION_1SHOT_MASK,
-							 1);
+				ret = max31827_update_bits(st->client,
+
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_1SHOT_MASK,
+							   1);
   				if (ret) {
   					mutex_unlock(&st->lock);
   					return ret;
@@ -246,7 +293,8 @@ static int max31827_read(struct device *dev,
enum hwmon_sensor_types type,
   			    st->update_interval == 125)
   				usleep_range(15000, 20000);

-			ret = regmap_read(st->regmap,
MAX31827_T_REG, &uval);
+			ret = max31827_reg_read(st->client,
MAX31827_T_REG,
+						&uval);

   			mutex_unlock(&st->lock);

@@ -257,23 +305,26 @@ static int max31827_read(struct device *dev,
enum hwmon_sensor_types type,

   			break;
   		case hwmon_temp_max:
-			ret = regmap_read(st->regmap,
MAX31827_TH_REG, &uval);
+			ret = max31827_reg_read(st->client,
MAX31827_TH_REG,
+						&uval);
   			if (ret)
   				break;

   			*val = MAX31827_16_BIT_TO_M_DGR(uval);
   			break;
   		case hwmon_temp_max_hyst:
-			ret = regmap_read(st->regmap,
MAX31827_TH_HYST_REG,
-					  &uval);
+			ret = max31827_reg_read(st->client,
+
	MAX31827_TH_HYST_REG,
+						&uval);
   			if (ret)
   				break;

   			*val = MAX31827_16_BIT_TO_M_DGR(uval);
   			break;
   		case hwmon_temp_max_alarm:
-			ret = regmap_read(st->regmap,
-
MAX31827_CONFIGURATION_REG, &uval);
+			ret = max31827_reg_read(st->client,
+
	MAX31827_CONFIGURATION_REG,
+						&uval);
   			if (ret)
   				break;

@@ -281,23 +332,25 @@ static int max31827_read(struct device *dev,
enum hwmon_sensor_types type,
   					 uval);
   			break;
   		case hwmon_temp_min:
-			ret = regmap_read(st->regmap,
MAX31827_TL_REG, &uval);
+			ret = max31827_reg_read(st->client,
MAX31827_TL_REG,
+						&uval);
   			if (ret)
   				break;

   			*val = MAX31827_16_BIT_TO_M_DGR(uval);
   			break;
   		case hwmon_temp_min_hyst:
-			ret = regmap_read(st->regmap,
MAX31827_TL_HYST_REG,
-					  &uval);
+			ret = max31827_reg_read(st->client,
MAX31827_TL_HYST_REG,
+						&uval);
   			if (ret)
   				break;

   			*val = MAX31827_16_BIT_TO_M_DGR(uval);
   			break;
   		case hwmon_temp_min_alarm:
-			ret = regmap_read(st->regmap,
-
MAX31827_CONFIGURATION_REG, &uval);
+			ret = max31827_reg_read(st->client,
+
	MAX31827_CONFIGURATION_REG,
+						&uval);
   			if (ret)
   				break;

@@ -313,8 +366,9 @@ static int max31827_read(struct device *dev,
enum
hwmon_sensor_types type,

   	case hwmon_chip:
   		if (attr == hwmon_chip_update_interval) {
-			ret = regmap_read(st->regmap,
-
MAX31827_CONFIGURATION_REG, &uval);
+			ret = max31827_reg_read(st->client,
+
	MAX31827_CONFIGURATION_REG,
+						&uval);
   			if (ret)
   				break;

@@ -355,11 +409,11 @@ static int max31827_write(struct device *dev,
enum hwmon_sensor_types type,

   			st->enable = val;

-			ret = regmap_update_bits(st->regmap,
-
MAX31827_CONFIGURATION_REG,
-
MAX31827_CONFIGURATION_1SHOT_MASK |
-
MAX31827_CONFIGURATION_CNV_RATE_MASK,
-
MAX31827_DEVICE_ENABLE(val));
+			ret = max31827_update_bits(st->client,
+
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_1SHOT_MASK |
+
MAX31827_CONFIGURATION_CNV_RATE_MASK,
+
MAX31827_DEVICE_ENABLE(val));

   			mutex_unlock(&st->lock);

@@ -402,10 +456,10 @@ static int max31827_write(struct device *dev,
enum hwmon_sensor_types type,
   			res =
FIELD_PREP(MAX31827_CONFIGURATION_CNV_RATE_MASK,
   					 res);

-			ret = regmap_update_bits(st->regmap,
-
MAX31827_CONFIGURATION_REG,
-
MAX31827_CONFIGURATION_CNV_RATE_MASK,
-						 res);
+			ret = max31827_update_bits(st->client,
+
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_CNV_RATE_MASK,
+						   res);
   			if (ret)
   				return ret;

@@ -425,10 +479,10 @@ static ssize_t temp1_resolution_show(struct
device *dev,
   				     char *buf)
   {
   	struct max31827_state *st = dev_get_drvdata(dev);
-	unsigned int val;
+	u16 val;
   	int ret;

-	ret = regmap_read(st->regmap,
MAX31827_CONFIGURATION_REG, &val);
+	ret = max31827_reg_read(st->client,
MAX31827_CONFIGURATION_REG,
+&val);
   	if (ret)
   		return ret;

@@ -473,10 +527,63 @@ static ssize_t temp1_resolution_store(struct
device *dev,
   	return ret ? ret : count;
   }

+static ssize_t pec_show(struct device *dev, struct device_attribute
*devattr,
+			char *buf)
+{
+	struct max31827_state *st = dev_get_drvdata(dev);
+	struct i2c_client *client = st->client;
+
+	return scnprintf(buf, PAGE_SIZE, "%d\n", !!(client->flags &
+I2C_CLIENT_PEC)); }
+
+static ssize_t pec_store(struct device *dev, struct device_attribute
*devattr,
+			 const char *buf, size_t count)
+{
+	struct max31827_state *st = dev_get_drvdata(dev);
+	struct i2c_client *client = st->client;
+	unsigned int val;
+	u16 val2;
+	int err;
+
+	err = kstrtouint(buf, 10, &val);
+	if (err < 0)
+		return err;
+
+	val2 = FIELD_PREP(MAX31827_CONFIGURATION_PEC_EN_MASK,
val);
+
+	if (err)
+		return err;
+
+	switch (val) {
+	case 0:
+		client->flags &= ~I2C_CLIENT_PEC;
+		err = max31827_update_bits(client,
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_PEC_EN_MASK,
+					   val2);
+		if (err)
+			return err;
+		break;
+	case 1:
+		err = max31827_update_bits(client,
MAX31827_CONFIGURATION_REG,
+
MAX31827_CONFIGURATION_PEC_EN_MASK,
+					   val2);
+		if (err)
+			return err;
+		client->flags |= I2C_CLIENT_PEC;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return count;
+}
+
   static DEVICE_ATTR_RW(temp1_resolution);
+static DEVICE_ATTR_RW(pec);

   static struct attribute *max31827_attrs[] = {
   	&dev_attr_temp1_resolution.attr,
+	&dev_attr_pec.attr,
   	NULL
   };
   ATTRIBUTE_GROUPS(max31827);
@@ -489,9 +596,9 @@ static const struct i2c_device_id
max31827_i2c_ids[] = {
   };
   MODULE_DEVICE_TABLE(i2c, max31827_i2c_ids);

-static int max31827_init_client(struct max31827_state *st,
-				struct device *dev)
+static int max31827_init_client(struct max31827_state *st)
   {
+	struct device *dev = &st->client->dev;

Now we are absolutely down to personal preference changes.
I am not at all inclined to accept such changes, sorry.

Including such changes means I'll have to put extra scrutiny on your patch
submissions in the future to ensure that you don't try to sneak in similar
changes, which I find quite frustrating. Is that really necessary ?

Guenter


Sorry for this! I thought, if I am adding client to the private structure I might as well delete the second parameter of init_client, because I can easily retrieve the device structure from client. I added this line so that the changes to the code are kept to a minimum.


You are making unnecessary changes (several unsigned int -> u16
plus this one), and claim this would be "so that the changes to
the code are kept to a minimum". Really ? How does making
unnecessary changes keep the changes to the code to a minimum ?

Guenter








