On 09/09/2022 16:44, Martin Povišer wrote:
On 9. 9. 2022, at 17:40, Richard Fitzgerald <rf@xxxxxxxxxxxxxxxxxxxxx> wrote:
On 09/09/2022 14:53, Martin Povišer wrote:
+static int cs42l42_i2c_probe(struct i2c_client *i2c_client)
+{
+ struct device *dev = &i2c_client->dev;
+ struct cs42l42_private *cs42l42;
+ struct regmap *regmap;
+ int ret;
+
+ cs42l42 = devm_kzalloc(dev, sizeof(*cs42l42), GFP_KERNEL);
+ if (!cs42l42)
+ return -ENOMEM;
+
+ regmap = devm_regmap_init_i2c(i2c_client, &cs42l42_regmap);
+ if (IS_ERR(regmap)) {
+ ret = PTR_ERR(regmap);
+ dev_err(&i2c_client->dev, "regmap_init() failed: %d\n", ret);
+ return ret;
+ }
+
+ cs42l42->dev = dev;
+ cs42l42->regmap = regmap;
+ cs42l42->irq = i2c_client->irq;
+
+ ret = cs42l42_common_probe(cs42l42, &cs42l42_soc_component, &cs42l42_dai);
+ if (ret)
+ return ret;
+
+ ret = cs42l42_init(cs42l42);
+ if (ret)
+ cs42l42_common_remove(cs42l42);
This introduces a bug that regulator_bulk_disable() is called
twice if there is an error.
cs42l42_init() was supposed to clean up if it returns an error, which
it nearly does, but my original patch is missing the call to free_irq()
in the error paths of cs42l42_init().
Ah! I didn’t inspect it closely enough then, I only ran into the missing
free_irq.
Yes, that's a bug. I just put a comment on the patch that introduced it.
When I split probe() into two, I accidentally missed out those two lines
to call free_irq().
Martin
