STM32L4 I2C read polling hang on NACK in stm32_i2c_msg_read()

[eqvinox]

When trying to do an I2C read that is getting NACK'd in the address phase, stm32_i2c_msg_read() will spin forever because RXNE will never be set. This causes the innermost loop to spin forever.

Similar to #8915, #7626 and commit 81cfdec, but this issue is in polling mode, not interrupt.

[nashif]

@erwango will you take care of this?

[ydamigos]

diff --git a/drivers/i2c/i2c_ll_stm32_v2.c b/drivers/i2c/i2c_ll_stm32_v2.c                                        
index 43a8a9113..2a2f5ba15 100644
--- a/drivers/i2c/i2c_ll_stm32_v2.c
+++ b/drivers/i2c/i2c_ll_stm32_v2.c
@@ -535,7 +535,9 @@ int stm32_i2c_msg_read(struct device *dev, struct i2c_msg *msg,
        len = msg->len;
        while (len) {
                while (!LL_I2C_IsActiveFlag_RXNE(i2c)) {
-                       ;
+                       if (LL_I2C_IsActiveFlag_NACK(i2c)) {
+                               goto error;
+                       }
                }
 
                *buf = LL_I2C_ReceiveData8(i2c);
@@ -546,6 +548,11 @@ int stm32_i2c_msg_read(struct device *dev, struct i2c_msg *msg,
        msg_done(dev, msg->flags);
 
        return 0;
+error:
+       LL_I2C_ClearFlag_NACK(i2c);
+       SYS_LOG_DBG("%s: NACK", __func__);
+
+       return -EIO;
 }
 #endif

@eqvinox Does the above patch fixes it?

[eqvinox]

@ydamigos sorry I'll probably need about a week or so until I can test this, I'm out sick 🤒

[psidhu]

@ydamigos I got into this state once. Your patch solved that issue.