Skip to contentSkip to conversation

hwmon: (pmbus/ltc4286) two patches about device matching

PATCH Applied Uwe Kleine-König 2 participants 8 messages Jul 10, 2024 – Jul 11, 2024

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover hwmon: (pmbus/ltc4286) two patches about device matching

Hello,

while working on my quest to improve i2c_device_id I noticed some minor strange things in the pmbus/ltc4286 driver. The first patch just documents the first peculiarity found, so the next reader doesn't wonder if this is done on purpose (and the patch gives the opportunity to notice if this behaviour is unintended). The second drops the other peculiarity I identified.

Best regards
Uwe

Uwe Kleine-König (2):
hwmon: (pmbus/ltc4286) Improve device matching
hwmon: (pmbus/ltc4286) Drop unused i2c device ids

 drivers/hwmon/pmbus/ltc4286.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

base-commit: 82d01fe6ee52086035b201cfa1410a3b04384257

Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/2 hwmon: (pmbus/ltc4286) Improve device matching

The devices supported by this driver report the model name in their register space. The way this is evaluated allows longer strings than the driver's model list. Document this behaviour in a code comment to lessen the surprise for the next reader.

Additionally emit the reported model name in case of a mismatch.

---
 drivers/hwmon/pmbus/ltc4286.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +9 -3
index 9e7ceeb7e789..2e5532300eff 100644
@@ -95,13 +95,19 @@ static int ltc4286_probe(struct i2c_client *client)
 				     "Failed to read manufacturer model\n");
 	}
 
-	for (mid = ltc4286_id; mid->name[0]; mid++) {
+	for (mid = ltc4286_id; mid->name[0]; mid++)
+		/*
+		 * Note that by limiting the comparison to strlen(mid->name)
+		 * chars, the device reporting "lTc4286chocolade" is accepted,
+		 * too.
+		 */
 		if (!strncasecmp(mid->name, block_buffer, strlen(mid->name)))
 			break;
-	}
+
 	if (!mid->name[0])
 		return dev_err_probe(&client->dev, -ENODEV,
-				     "Unsupported device\n");
+				     "Unsupported device (reported: \"%*pE\")\n",
+				     ret, block_buffer);
 
 	if (of_property_read_u32(client->dev.of_node,
 				 "shunt-resistor-micro-ohms", &rsense))
Signature · 1 line
2.43.0
maintainer
#2 lore raw

Re: [PATCH 1/2] hwmon: (pmbus/ltc4286) Improve device matching

Quoted 17 lines · Uwe Kleine-König

The devices supported by this driver report the model name in their register space. The way this is evaluated allows longer strings than the driver's model list. Document this behaviour in a code comment to lessen the surprise for the next reader.

Additionally emit the reported model name in case of a mismatch.

---
  drivers/hwmon/pmbus/ltc4286.c | 12 +++++++++---
  1 file changed, 9 insertions(+), 3 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +0 -0
index 9e7ceeb7e789..2e5532300eff 100644

drivers/hwmon/pmbus/ltc4286.c static int ltc4286_probe(struct i2c_client *client) · line 103

Quoted 10 lines
drivers/hwmon/pmbus/ltc4286.c +6 -1
@@ -95,13 +95,19 @@ static int ltc4286_probe(struct i2c_client *client)
  				     "Failed to read manufacturer model\n");
  	}
 
-	for (mid = ltc4286_id; mid->name[0]; mid++) {
+	for (mid = ltc4286_id; mid->name[0]; mid++)
+		/*
+		 * Note that by limiting the comparison to strlen(mid->name)
+		 * chars, the device reporting "lTc4286chocolade" is accepted,
+		 * too.
+		 */

This is misleading; the desired match is LTC4286 and all its variants (LTC4286[A-Z] and whatever else the vendor can come up with), i.e., it is supposed to include all device variants, and ignoring case since it is irrelevant. Referring to the odd string just makes that look unnecessarily bad. I am not going to apply this patch, sorry.

Guenter

#3 lore raw

Re: [PATCH 1/2] hwmon: (pmbus/ltc4286) Improve device matching

Hello Guenter,

Quoted 24 lines · Guenter Roeck
Quoted 23 lines · Uwe Kleine-König

The devices supported by this driver report the model name in their register space. The way this is evaluated allows longer strings than the driver's model list. Document this behaviour in a code comment to lessen the surprise for the next reader.

Additionally emit the reported model name in case of a mismatch.

---
  drivers/hwmon/pmbus/ltc4286.c | 12 +++++++++---
  1 file changed, 9 insertions(+), 3 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +3 -1
index 9e7ceeb7e789..2e5532300eff 100644
@@ -95,13 +95,19 @@ static int ltc4286_probe(struct i2c_client *client)
  				     "Failed to read manufacturer model\n");
  	}
-	for (mid = ltc4286_id; mid->name[0]; mid++) {
+	for (mid = ltc4286_id; mid->name[0]; mid++)
+		/*
+		 * Note that by limiting the comparison to strlen(mid->name)
Quoted 8 lines

drivers/hwmon/pmbus/ltc4286.c static int ltc4286_probe(struct i2c_client *client) · line 103

Quoted 3 lines
drivers/hwmon/pmbus/ltc4286.c +3 -0
+		 * chars, the device reporting "lTc4286chocolade" is accepted,
+		 * too.
+		 */

This is misleading; the desired match is LTC4286 and all its variants (LTC4286[A-Z] and whatever else the vendor can come up with), i.e., it is supposed to include all device variants, and ignoring case since it is irrelevant. Referring to the odd string just makes that look unnecessarily bad. I am not going to apply this patch, sorry.

You're quite an optimist, expecting "whatever the vendor can come up with" but nothing bad :-)

Anyhow, what about updating the comment to read:

Note that by limiting the comparison to strlen(mid->name) chars, matching for devices that report their model with a variant suffix is supported.

While looking at the code again, I spotted a (theoretic) bug: Given that block_buffer isn't initialized at function entry, it might well contain "LTC4286something" (which might even be realistic if the driver just probed on a different bus?). Now if i2c_smbus_read_block_data(... PMBUS_MFR_MODEL, ...) returned something between 0 and 6, we're looking at bytes that didn't come from the block read.

Best regards
Uwe

maintainer
#4 lore raw

Re: [PATCH 1/2] hwmon: (pmbus/ltc4286) Improve device matching

Quoted 36 lines · Uwe Kleine-König

Hello Guenter,

Quoted 24 lines · Guenter Roeck
Quoted 23 lines · Uwe Kleine-König

The devices supported by this driver report the model name in their register space. The way this is evaluated allows longer strings than the driver's model list. Document this behaviour in a code comment to lessen the surprise for the next reader.

Additionally emit the reported model name in case of a mismatch.

---
   drivers/hwmon/pmbus/ltc4286.c | 12 +++++++++---
   1 file changed, 9 insertions(+), 3 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +3 -1
index 9e7ceeb7e789..2e5532300eff 100644
@@ -95,13 +95,19 @@ static int ltc4286_probe(struct i2c_client *client)
   				     "Failed to read manufacturer model\n");
   	}
-	for (mid = ltc4286_id; mid->name[0]; mid++) {
+	for (mid = ltc4286_id; mid->name[0]; mid++)
+		/*
+		 * Note that by limiting the comparison to strlen(mid->name)
Quoted 8 lines

drivers/hwmon/pmbus/ltc4286.c static int ltc4286_probe(struct i2c_client *client) · line 103

Quoted 3 lines
drivers/hwmon/pmbus/ltc4286.c +3 -0
+		 * chars, the device reporting "lTc4286chocolade" is accepted,
+		 * too.
+		 */

This is misleading; the desired match is LTC4286 and all its variants (LTC4286[A-Z] and whatever else the vendor can come up with), i.e., it is supposed to include all device variants, and ignoring case since it is irrelevant. Referring to the odd string just makes that look unnecessarily bad. I am not going to apply this patch, sorry.

Quoted 3 lines

You're quite an optimist, expecting "whatever the vendor can come up with" but nothing bad :-)

"optimist" is relative. A perfectly valid alternative would be to _not_ do any testing at all. After all, this is not a detect function, this is the probe function, which should only be called _after_ the chip has been identified.

Since the model number is not used for anything but extra validation, one might as well argue that the validation is unnecessary and can or should be dropped to reduce boot time. Of course, given the vagueness of the PMBus specification, that might result in fatal consequences if the wrong chip is instantiated, so I think that validation does make sense, and I suggest to add it for all PMBus drivers. However, one can overdo it (and not all drivers do it).

Quoted 6 lines

Anyhow, what about updating the comment to read:

Note that by limiting the comparison to strlen(mid->name) chars, matching for devices that report their model with a variant suffix is supported.

Quoted 7 lines

While looking at the code again, I spotted a (theoretic) bug: Given that block_buffer isn't initialized at function entry, it might well contain "LTC4286something" (which might even be realistic if the driver just probed on a different bus?). Now if i2c_smbus_read_block_data(... PMBUS_MFR_MODEL, ...) returned something between 0 and 6, we're looking at bytes that didn't come from the block read.

Yes, I would agree that a check ensuring that ret >= 7 would make sense.

Thanks,
Guenter

#5 lore raw

Re: [PATCH 1/2] hwmon: (pmbus/ltc4286) Improve device matching

Hello Guenter,

Quoted 43 lines · Guenter Roeck
Quoted 34 lines · Uwe Kleine-König
Quoted 24 lines · Guenter Roeck
Quoted 23 lines · Uwe Kleine-König

The devices supported by this driver report the model name in their register space. The way this is evaluated allows longer strings than the driver's model list. Document this behaviour in a code comment to lessen the surprise for the next reader.

Additionally emit the reported model name in case of a mismatch.

---
   drivers/hwmon/pmbus/ltc4286.c | 12 +++++++++---
   1 file changed, 9 insertions(+), 3 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +3 -1
index 9e7ceeb7e789..2e5532300eff 100644
@@ -95,13 +95,19 @@ static int ltc4286_probe(struct i2c_client *client)
   				     "Failed to read manufacturer model\n");
   	}
-	for (mid = ltc4286_id; mid->name[0]; mid++) {
+	for (mid = ltc4286_id; mid->name[0]; mid++)
+		/*
+		 * Note that by limiting the comparison to strlen(mid->name)
Quoted 8 lines

drivers/hwmon/pmbus/ltc4286.c static int ltc4286_probe(struct i2c_client *client) · line 103

Quoted 3 lines
drivers/hwmon/pmbus/ltc4286.c +3 -0
+		 * chars, the device reporting "lTc4286chocolade" is accepted,
+		 * too.
+		 */

This is misleading; the desired match is LTC4286 and all its variants (LTC4286[A-Z] and whatever else the vendor can come up with), i.e., it is supposed to include all device variants, and ignoring case since it is irrelevant. Referring to the odd string just makes that look unnecessarily bad. I am not going to apply this patch, sorry.

Quoted 3 lines

You're quite an optimist, expecting "whatever the vendor can come up with" but nothing bad :-)

"optimist" is relative. A perfectly valid alternative would be to _not_ do any testing at all. After all, this is not a detect function, this is the probe function, which should only be called _after_ the chip has been identified.

Quoted 6 lines

Since the model number is not used for anything but extra validation, one might as well argue that the validation is unnecessary and can or should be dropped to reduce boot time. Of course, given the vagueness of the PMBus specification, that might result in fatal consequences if the wrong chip is instantiated, so I think that validation does make sense, and I suggest to add it for all PMBus drivers. However, one can overdo it (and not all drivers do it).

+1 for a generic check in generic code.

One could also argue that it's an error if the device was declared to be a ltc4286 but reports LTC4287A in its PMBUS_MFR_MODEL register. So something like:

	mid = i2c_client_get_device_id(client);

would make sense, too. (There is a corner case that the driver is not bound via the entries in the driver's .id_table, not sure how relevant this is.)

Quoted 14 lines
Quoted 6 lines

Anyhow, what about updating the comment to read:

Note that by limiting the comparison to strlen(mid->name) chars, matching for devices that report their model with a variant suffix is supported.

Quoted 7 lines

While looking at the code again, I spotted a (theoretic) bug: Given that block_buffer isn't initialized at function entry, it might well contain "LTC4286something" (which might even be realistic if the driver just probed on a different bus?). Now if i2c_smbus_read_block_data(... PMBUS_MFR_MODEL, ...) returned something between 0 and 6, we're looking at bytes that didn't come from the block read.

Quoted 1 line

Yes, I would agree that a check ensuring that ret >= 7 would make sense.

alternatively do

	block_buffer[ret] = '\0';

before the comparison.

To be honest, patch #2 was my focus and I don't have a pmbus device. So I'll drop this topic and let you (or someone else) handle the action items arising from this discussion.

Best regards
Uwe

#6 lore raw

PATCH 2/2 hwmon: (pmbus/ltc4286) Drop unused i2c device ids

In reply to Uwe Kleine-König

The driver doesn't make use of the different numbers assigned to the different devices. So drop them.

---
 drivers/hwmon/pmbus/ltc4286.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/hwmon/pmbus/ltc4286.c +2 -2
index 2e5532300eff..28aa211700fe 100644
@@ -58,8 +58,8 @@ static struct pmbus_driver_info ltc4286_info = {
 };
 
 static const struct i2c_device_id ltc4286_id[] = {
-	{ "ltc4286", 0 },
-	{ "ltc4287", 1 },
+	{ "ltc4286", },
+	{ "ltc4287", },
 	{}
 };
 MODULE_DEVICE_TABLE(i2c, ltc4286_id);
Signature · 1 line
2.43.0
maintainer Applied
#7 lore raw

Re: [PATCH 2/2] hwmon: (pmbus/ltc4286) Drop unused i2c device ids

Quoted 4 lines · Uwe Kleine-König

The driver doesn't make use of the different numbers assigned to the different devices. So drop them.

Applied.

Thanks,
Guenter