Skip to contentSkip to conversation

max17042_battery: Support regmap to access device's registers

DISCUSSION Dan Carpenter 2 participants 3 messages Aug 14, 2015 – Aug 17, 2015

lore
Share & tools
lore
#0 lore raw

Hello Jonghwa Lee,

The patch 39e7213edc4f: "max17042_battery: Support regmap to access device's registers" from Oct 25, 2013, leads to the following static checker warning:

	drivers/power/max17042_battery.c:501 max17042_init_model()
	warn: passing casted pointer 'temp_data' to 'max17042_model_data_compare()' 32 vs 16.

drivers/power/max17042_battery.c
469 static inline int max17042_model_data_compare(struct max17042_chip *chip,
470 u16 *data1, u16 *data2, int size)
471 {
472 int i;
473

   474          if (memcmp(data1, data2, size)) {
                                         ^^^^

data2 is size * sizeof(u32) bytes long so we are only checking the first quarter of the buffer. data1 is in size of u16s so I wonder how this memcmp() works at all?

475 dev_err(&chip->client->dev, "%s compare failed\n", __func__);
476 for (i = 0; i < size; i++)

                                        ^^^^

Here size is in terms of u16 elements. The original code mixed u16 units with byte units so it was buggy, but now we are using u32 units so it's even worse.

   477                          dev_info(&chip->client->dev, "0x%x, 0x%x",
   478                                  data1[i], data2[i]);
   479                  dev_info(&chip->client->dev, "\n");
   480                  return -EINVAL;
   481          }
   482          return 0;
   483  }
   484  
   485  static int max17042_init_model(struct max17042_chip *chip)
   486  {
   487          int ret;
   488          int table_size = ARRAY_SIZE(chip->pdata->config_data->cell_char_tbl);

data1 is any array of u16s.

   489          u32 *temp_data;
   490  
   491          temp_data = kcalloc(table_size, sizeof(*temp_data), GFP_KERNEL);

Here "size" (table_size) is in units of u32 elements.

492 if (!temp_data)
493 return -ENOMEM;
494
495 max10742_unlock_model(chip);
496 max17042_write_model_data(chip, MAX17042_MODELChrTbl,
497 table_size);
498 max17042_read_model_data(chip, MAX17042_MODELChrTbl, temp_data,

   499                                  table_size);
                                        ^^^^^^^^^^

Here "size" (table_size) is in units of u32 elements.

500
501 ret = max17042_model_data_compare(

   502                  chip,
   503                  chip->pdata->config_data->cell_char_tbl,
   504                  (u16 *)temp_data,
                        ^^^^^^^^^^^^^^^^

This cast is ugly and causes a static checker warning.

   505                  table_size);
   506  
   507          max10742_lock_model(chip);
   508          kfree(temp_data);
   509  
   510          return ret;
   511  }

regards,
dan carpenter

#1 lore raw

2015-08-14 18:55 GMT+09:00 Dan Carpenter <[email protected]>:

Quoted 69 lines

Hello Jonghwa Lee,

The patch 39e7213edc4f: "max17042_battery: Support regmap to access device's registers" from Oct 25, 2013, leads to the following static checker warning:

drivers/power/max17042_battery.c:501 max17042_init_model()
warn: passing casted pointer 'temp_data' to 'max17042_model_data_compare()' 32 vs 16.

drivers/power/max17042_battery.c
469 static inline int max17042_model_data_compare(struct max17042_chip *chip,
470 u16 *data1, u16 *data2, int size)

   471  {
   472          int i;
   473
   474          if (memcmp(data1, data2, size)) {
                                         ^^^^

data2 is size * sizeof(u32) bytes long so we are only checking the first quarter of the buffer. data1 is in size of u16s so I wonder how this memcmp() works at all?

   475                  dev_err(&chip->client->dev, "%s compare failed\n", __func__);
   476                  for (i = 0; i < size; i++)
                                        ^^^^

Here size is in terms of u16 elements. The original code mixed u16 units with byte units so it was buggy, but now we are using u32 units so it's even worse.

   477                          dev_info(&chip->client->dev, "0x%x, 0x%x",
   478                                  data1[i], data2[i]);
   479                  dev_info(&chip->client->dev, "\n");
   480                  return -EINVAL;
   481          }
   482          return 0;
   483  }
   484
   485  static int max17042_init_model(struct max17042_chip *chip)
   486  {
   487          int ret;
   488          int table_size = ARRAY_SIZE(chip->pdata->config_data->cell_char_tbl);

data1 is any array of u16s.

   489          u32 *temp_data;
   490
   491          temp_data = kcalloc(table_size, sizeof(*temp_data), GFP_KERNEL);

Here "size" (table_size) is in units of u32 elements.

   492          if (!temp_data)
   493                  return -ENOMEM;
   494
   495          max10742_unlock_model(chip);
   496          max17042_write_model_data(chip, MAX17042_MODELChrTbl,
   497                                  table_size);
   498          max17042_read_model_data(chip, MAX17042_MODELChrTbl, temp_data,
   499                                  table_size);
                                        ^^^^^^^^^^

Here "size" (table_size) is in units of u32 elements.

   500
   501          ret = max17042_model_data_compare(
   502                  chip,
   503                  chip->pdata->config_data->cell_char_tbl,
   504                  (u16 *)temp_data,
                        ^^^^^^^^^^^^^^^^

This cast is ugly and causes a static checker warning.

Quoted 7 lines
   505                  table_size);
   506
   507          max10742_lock_model(chip);
   508          kfree(temp_data);
   509
   510          return ret;
   511  }

The cast is indeed ugly but I think that in each loop the u16 pointers are used so this should work. regmap also uses 16-bit configuration. memcpy operates on bytes, not on words. The 'temp_data' (which is size of u32*table_size) is just unnecessarily big, actually two times bigger than needed.

Where exactly do you see an error (except wasted space)?

Best regards,
Krzysztof

#2 lore raw

The static checker is just complaining about the cast... I have looked at it again and this code is definitely buggy.

max17042_read_model_data() reads u32s into tmp_data. ->cell_char_tbl is an array with 48 elements of u16.

We later assume that cell_char_tbl and temp_data have 48 bytes each when we use memcmp() to compare them. This only compares the first half of ->cell_char_tbl with the first quarter of tmp_data[].

Then we assume that they hold u16s when we print out the warning message.

regards,
dan carpetner