Skip to contentSkip to conversation

mfd: lp87565: Handle optional reset pin

bug report Dan Carpenter 3 participants 7 messages Nov 17, 2021 – Nov 18, 2021 Kernel janitors

lore
Share & tools
lore
#0 lore raw

Hello Luca Ceresoli,

The patch 50e4d7a2a667: "mfd: lp87565: Handle optional reset pin" from Feb 26, 2021, leads to the following Smatch static checker warning:

	drivers/mfd/lp87565.c:76 lp87565_probe()
	warn: 'lp87565->reset_gpio' could be an error pointer

drivers/mfd/lp87565.c

    46 static int lp87565_probe(struct i2c_client *client,
    47                          const struct i2c_device_id *ids)
    48 {
    49         struct lp87565 *lp87565;
    50         const struct of_device_id *of_id;
    51         int ret;
    52         unsigned int otpid;
    53 
    54         lp87565 = devm_kzalloc(&client->dev, sizeof(*lp87565), GFP_KERNEL);
    55         if (!lp87565)
    56                 return -ENOMEM;
    57 
    58         lp87565->dev = &client->dev;
    59 
    60         lp87565->regmap = devm_regmap_init_i2c(client, &lp87565_regmap_config);
    61         if (IS_ERR(lp87565->regmap)) {
    62                 ret = PTR_ERR(lp87565->regmap);
    63                 dev_err(lp87565->dev,
    64                         "Failed to initialize register map: %d\n", ret);
    65                 return ret;
    66         }
    67 
    68         lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
    69                                                       GPIOD_OUT_LOW);
    70         if (IS_ERR(lp87565->reset_gpio)) {
    71                 ret = PTR_ERR(lp87565->reset_gpio);
    72                 if (ret == -EPROBE_DEFER)
    73                         return ret;

Only "ret = -EPROBE_DEFER" is handled. Other error pointer will lead to a crash.

    74         }
    75 
--> 76         if (lp87565->reset_gpio) {
    77                 gpiod_set_value_cansleep(lp87565->reset_gpio, 1);
    78                 /* The minimum assertion time is undocumented, just guess */
    79                 usleep_range(2000, 4000);
    80 
    81                 gpiod_set_value_cansleep(lp87565->reset_gpio, 0);
    82                 /* Min 1.2 ms before first I2C transaction */
    83                 usleep_range(1500, 3000);
    84         }
    85 
    86         ret = regmap_read(lp87565->regmap, LP87565_REG_OTP_REV, &otpid);
    87         if (ret) {
    88                 dev_err(lp87565->dev, "Failed to read OTP ID\n");
    89                 return ret;
    90         }
    91 
    92         lp87565->rev = otpid & LP87565_OTP_REV_OTP_ID;
    93 
    94         of_id = of_match_device(of_lp87565_match_table, &client->dev);
    95         if (of_id)
    96                 lp87565->dev_type = (enum lp87565_device_type)of_id->data;
    97 
    98         i2c_set_clientdata(client, lp87565);
    99 
    100         return devm_mfd_add_devices(lp87565->dev, PLATFORM_DEVID_AUTO,
    101                                     lp87565_cells, ARRAY_SIZE(lp87565_cells),
    102                                     NULL, 0, NULL);
    103 }

regards,
dan carpenter

#1 lore raw
Quoted 39 lines · Dan Carpenter

Hello Luca Ceresoli,

The patch 50e4d7a2a667: "mfd: lp87565: Handle optional reset pin" from Feb 26, 2021, leads to the following Smatch static checker warning:

	drivers/mfd/lp87565.c:76 lp87565_probe()
	warn: 'lp87565->reset_gpio' could be an error pointer

drivers/mfd/lp87565.c

    46 static int lp87565_probe(struct i2c_client *client,
    47                          const struct i2c_device_id *ids)
    48 {
    49         struct lp87565 *lp87565;
    50         const struct of_device_id *of_id;
    51         int ret;
    52         unsigned int otpid;
    53 
    54         lp87565 = devm_kzalloc(&client->dev, sizeof(*lp87565), GFP_KERNEL);
    55         if (!lp87565)
    56                 return -ENOMEM;
    57 
    58         lp87565->dev = &client->dev;
    59 
    60         lp87565->regmap = devm_regmap_init_i2c(client, &lp87565_regmap_config);
    61         if (IS_ERR(lp87565->regmap)) {
    62                 ret = PTR_ERR(lp87565->regmap);
    63                 dev_err(lp87565->dev,
    64                         "Failed to initialize register map: %d\n", ret);
    65                 return ret;
    66         }
    67 
    68         lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
    69                                                       GPIOD_OUT_LOW);
    70         if (IS_ERR(lp87565->reset_gpio)) {
    71                 ret = PTR_ERR(lp87565->reset_gpio);
    72                 if (ret == -EPROBE_DEFER)
    73                         return ret;
Quoted 6 lines

Only "ret = -EPROBE_DEFER" is handled. Other error pointer will lead to a crash.

    74         }
    75 

--> 76 if (lp87565->reset_gpio) {

I guess this should be:

    if (lp87565->reset_gpio >= 0)

If 0 is valid, or '>' if it's not.

Quoted 30 lines
    77                 gpiod_set_value_cansleep(lp87565->reset_gpio, 1);
    78                 /* The minimum assertion time is undocumented, just guess */
    79                 usleep_range(2000, 4000);
    80 
    81                 gpiod_set_value_cansleep(lp87565->reset_gpio, 0);
    82                 /* Min 1.2 ms before first I2C transaction */
    83                 usleep_range(1500, 3000);
    84         }
    85 
    86         ret = regmap_read(lp87565->regmap, LP87565_REG_OTP_REV, &otpid);
    87         if (ret) {
    88                 dev_err(lp87565->dev, "Failed to read OTP ID\n");
    89                 return ret;
    90         }
    91 
    92         lp87565->rev = otpid & LP87565_OTP_REV_OTP_ID;
    93 
    94         of_id = of_match_device(of_lp87565_match_table, &client->dev);
    95         if (of_id)
    96                 lp87565->dev_type = (enum lp87565_device_type)of_id->data;
    97 
    98         i2c_set_clientdata(client, lp87565);
    99 
    100         return devm_mfd_add_devices(lp87565->dev, PLATFORM_DEVID_AUTO,
    101                                     lp87565_cells, ARRAY_SIZE(lp87565_cells),
    102                                     NULL, 0, NULL);
    103 }

regards,
dan carpenter

Signature · 4 lines
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog
#2 lore raw
Quoted 48 lines · Lee Jones
Quoted 39 lines · Dan Carpenter

Hello Luca Ceresoli,

The patch 50e4d7a2a667: "mfd: lp87565: Handle optional reset pin" from Feb 26, 2021, leads to the following Smatch static checker warning:

	drivers/mfd/lp87565.c:76 lp87565_probe()
	warn: 'lp87565->reset_gpio' could be an error pointer

drivers/mfd/lp87565.c

    46 static int lp87565_probe(struct i2c_client *client,
    47                          const struct i2c_device_id *ids)
    48 {
    49         struct lp87565 *lp87565;
    50         const struct of_device_id *of_id;
    51         int ret;
    52         unsigned int otpid;
    53 
    54         lp87565 = devm_kzalloc(&client->dev, sizeof(*lp87565), GFP_KERNEL);
    55         if (!lp87565)
    56                 return -ENOMEM;
    57 
    58         lp87565->dev = &client->dev;
    59 
    60         lp87565->regmap = devm_regmap_init_i2c(client, &lp87565_regmap_config);
    61         if (IS_ERR(lp87565->regmap)) {
    62                 ret = PTR_ERR(lp87565->regmap);
    63                 dev_err(lp87565->dev,
    64                         "Failed to initialize register map: %d\n", ret);
    65                 return ret;
    66         }
    67 
    68         lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
    69                                                       GPIOD_OUT_LOW);
    70         if (IS_ERR(lp87565->reset_gpio)) {
    71                 ret = PTR_ERR(lp87565->reset_gpio);
    72                 if (ret == -EPROBE_DEFER)
    73                         return ret;
Quoted 6 lines

Only "ret = -EPROBE_DEFER" is handled. Other error pointer will lead to a crash.

    74         }
    75 

--> 76 if (lp87565->reset_gpio) {

Quoted 6 lines

I guess this should be:

    if (lp87565->reset_gpio >= 0)

If 0 is valid, or '>' if it's not.

lp87565->reset_gpio is a pointer so that's not going to work.

There are two options. The first is to just "return ret;" for all errors. The second option is to say something like:

	if (IS_ERR(lp87565->reset_gpio)) {
		ret = dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
				    "no GPIO error for ->reset_gpio");
		if (ret == -EPROBE_DEFER)
			return;
		lp87565->reset_gpio = NULL;
	}

regards,
dan carpenter

#3 lore raw
Quoted 49 lines · Dan Carpenter
Quoted 48 lines · Lee Jones
Quoted 45 lines · Dan Carpenter

Hello Luca Ceresoli,

The patch 50e4d7a2a667: "mfd: lp87565: Handle optional reset pin" from Feb 26, 2021, leads to the following Smatch static checker warning:

	drivers/mfd/lp87565.c:76 lp87565_probe()
	warn: 'lp87565->reset_gpio' could be an error pointer

drivers/mfd/lp87565.c

    46 static int lp87565_probe(struct i2c_client *client,
    47                          const struct i2c_device_id *ids)
    48 {
    49         struct lp87565 *lp87565;
    50         const struct of_device_id *of_id;
    51         int ret;
    52         unsigned int otpid;
    53 
    54         lp87565 = devm_kzalloc(&client->dev, sizeof(*lp87565), GFP_KERNEL);
    55         if (!lp87565)
    56                 return -ENOMEM;
    57 
    58         lp87565->dev = &client->dev;
    59 
    60         lp87565->regmap = devm_regmap_init_i2c(client, &lp87565_regmap_config);
    61         if (IS_ERR(lp87565->regmap)) {
    62                 ret = PTR_ERR(lp87565->regmap);
    63                 dev_err(lp87565->dev,
    64                         "Failed to initialize register map: %d\n", ret);
    65                 return ret;
    66         }
    67 
    68         lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
    69                                                       GPIOD_OUT_LOW);
    70         if (IS_ERR(lp87565->reset_gpio)) {
    71                 ret = PTR_ERR(lp87565->reset_gpio);
    72                 if (ret == -EPROBE_DEFER)
    73                         return ret;

Only "ret = -EPROBE_DEFER" is handled. Other error pointer will lead to a crash.

    74         }
    75 

--> 76 if (lp87565->reset_gpio) {

Quoted 8 lines
Quoted 6 lines

I guess this should be:

    if (lp87565->reset_gpio >= 0)

If 0 is valid, or '>' if it's not.

lp87565->reset_gpio is a pointer so that's not going to work.

Ah, I assumed this was a GPIO #. I missed the gpioD part.

Quoted 2 lines

There are two options. The first is to just "return ret;" for all errors.

No, don't do that. This GPIO is optional.

Quoted 9 lines

The second option is to say something like:

	if (IS_ERR(lp87565->reset_gpio)) {
		ret = dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
				    "no GPIO error for ->reset_gpio");
		if (ret == -EPROBE_DEFER)
			return;
		lp87565->reset_gpio = NULL;
	}

Works for me.

Or:

    if (!IS_ERR(lp87565->reset_gpio))

Or stick it in the else.

Signature · 4 lines
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog
#4 lore raw

Hi,

Quoted 77 lines · Lee Jones
Quoted 49 lines · Dan Carpenter
Quoted 48 lines · Lee Jones
Quoted 45 lines · Dan Carpenter

Hello Luca Ceresoli,

The patch 50e4d7a2a667: "mfd: lp87565: Handle optional reset pin" from Feb 26, 2021, leads to the following Smatch static checker warning:

	drivers/mfd/lp87565.c:76 lp87565_probe()
	warn: 'lp87565->reset_gpio' could be an error pointer

drivers/mfd/lp87565.c

    46 static int lp87565_probe(struct i2c_client *client,
    47                          const struct i2c_device_id *ids)
    48 {
    49         struct lp87565 *lp87565;
    50         const struct of_device_id *of_id;
    51         int ret;
    52         unsigned int otpid;
    53 
    54         lp87565 = devm_kzalloc(&client->dev, sizeof(*lp87565), GFP_KERNEL);
    55         if (!lp87565)
    56                 return -ENOMEM;
    57 
    58         lp87565->dev = &client->dev;
    59 
    60         lp87565->regmap = devm_regmap_init_i2c(client, &lp87565_regmap_config);
    61         if (IS_ERR(lp87565->regmap)) {
    62                 ret = PTR_ERR(lp87565->regmap);
    63                 dev_err(lp87565->dev,
    64                         "Failed to initialize register map: %d\n", ret);
    65                 return ret;
    66         }
    67 
    68         lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
    69                                                       GPIOD_OUT_LOW);
    70         if (IS_ERR(lp87565->reset_gpio)) {
    71                 ret = PTR_ERR(lp87565->reset_gpio);
    72                 if (ret == -EPROBE_DEFER)
    73                         return ret;

Only "ret = -EPROBE_DEFER" is handled. Other error pointer will lead to a crash.

    74         }
    75 

--> 76 if (lp87565->reset_gpio) {

Quoted 8 lines
Quoted 6 lines

I guess this should be:

    if (lp87565->reset_gpio >= 0)

If 0 is valid, or '>' if it's not.

lp87565->reset_gpio is a pointer so that's not going to work.

Ah, I assumed this was a GPIO #. I missed the gpioD part.

Quoted 2 lines

There are two options. The first is to just "return ret;" for all errors.

No, don't do that. This GPIO is optional.

Quoted 9 lines

The second option is to say something like:

	if (IS_ERR(lp87565->reset_gpio)) {
		ret = dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
				    "no GPIO error for ->reset_gpio");
		if (ret == -EPROBE_DEFER)
			return;
		lp87565->reset_gpio = NULL;
	}
Quoted 8 lines

Works for me.

Or:

    if (!IS_ERR(lp87565->reset_gpio))

Or stick it in the else.

As the author of the code to blame, I wrote this patch, but just needed a little time to test it before sending:

    lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
                                                  GPIOD_OUT_LOW);
    if (IS_ERR(lp87565->reset_gpio))
        return dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
                             "Failed getting reset GPIO");

    if (lp87565->reset_gpio) {
    ...

I prefer to exit on any error as it would be either -EPROBE_DEFER of a _real_ error (e.g. GPIO already in use). If there's no GPIO specified, then devm_gpiod_get_optional() returns NULL and libgpio ignores NULL pointers gracefully.

Would that work?

Signature · 1 line
Luca
#5 lore raw
Quoted 13 lines · Luca Ceresoli

As the author of the code to blame, I wrote this patch, but just needed a little time to test it before sending:

    lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
                                                  GPIOD_OUT_LOW);
    if (IS_ERR(lp87565->reset_gpio))
        return dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
                             "Failed getting reset GPIO");

    if (lp87565->reset_gpio) {
    ...
Quoted 6 lines

I prefer to exit on any error as it would be either -EPROBE_DEFER of a _real_ error (e.g. GPIO already in use). If there's no GPIO specified, then devm_gpiod_get_optional() returns NULL and libgpio ignores NULL pointers gracefully.

Would that work?

I generally prefer that as well, because to me optional means it's up to the user not that it's up to the kernel. But it depends on if the system can boot without it etc...

I guess in this case we know that no one was relying on the old behavior because that would have crashed so returning errors is safe.

regards,
dan carpenter

#6 lore raw

Hi,

Quoted 19 lines · Dan Carpenter
Quoted 18 lines · Luca Ceresoli

As the author of the code to blame, I wrote this patch, but just needed a little time to test it before sending:

    lp87565->reset_gpio = devm_gpiod_get_optional(lp87565->dev, "reset",
                                                  GPIOD_OUT_LOW);
    if (IS_ERR(lp87565->reset_gpio))
        return dev_err_probe(lp87565->dev, PTR_ERR(lp87565->reset_gpio),
                             "Failed getting reset GPIO");

    if (lp87565->reset_gpio) {
    ...

I prefer to exit on any error as it would be either -EPROBE_DEFER of a _real_ error (e.g. GPIO already in use). If there's no GPIO specified, then devm_gpiod_get_optional() returns NULL and libgpio ignores NULL pointers gracefully.

Quoted 8 lines
Quoted 1 line

Would that work?

I generally prefer that as well, because to me optional means it's up to the user not that it's up to the kernel. But it depends on if the system can boot without it etc...

I guess in this case we know that no one was relying on the old behavior because that would have crashed so returning errors is safe.

I sent a patch using this approach. I hope it's OK, otherwise I can send a v2.

Last but not least: thanks Dan for finding and reporting!

Signature · 1 line
Luca