Hi,

thanks for this driver!

+/**
+ * struct stm32f7_i2c_spec - private i2c specification timing
+ * @rate: I2C bus speed (Hz)
+ * @rate_min: 80% of I2C bus speed (Hz)
+ * @rate_max: 120% of I2C bus speed (Hz)

You would generate a clock which is higher than the requested one?
This is highly unusual. Any special reason?

+ * @fall_max: Max fall time of both SDA and SCL signals (ns)
+ * @rise_max: Max rise time of both SDA and SCL signals (ns)
+ * @hddat_min: Min data hold time (ns)
+ * @vddat_max: Max data valid time (ns)
+ * @sudat_min: Min data setup time (ns)
+ * @l_min: Min low period of the SCL clock (ns)
+ * @h_min: Min high period of the SCL clock (ns)
+ */
+static struct stm32f7_i2c_spec i2c_specs[] = {
+	[STM32_I2C_SPEED_STANDARD] = {
+		.rate = 100000,
+		.rate_min = 8000,

This is not 80%. Typo?

+		.rate_max = 120000,
+		.fall_max = 300,
+		.rise_max = 1000,
+		.hddat_min = 0,
+		.vddat_max = 3450,
+		.sudat_min = 250,
+		.l_min = 4700,
+		.h_min = 4000,
+	},

...

+	/*
+	 * Among Prescaler possibilities discovered above figures out SCL Low
+	 * and High Period. Provided:
+	 * - SCL Low Period has to be higher than Low Period of tehs SCL Clock

tehs?

+	 *   defined by I2C Specification. I2C Clock has to be lower than
+	 *   (SCL Low Period - Analog/Digital filters) / 4.
+	 * - SCL High Period has to be lower than High Period of the SCL Clock
+	 *   defined by I2C Specification
+	 * - I2C Clock has to be lower than SCL High Period
+	 */

...

+	/* NACK received */
+	if (status & STM32F7_I2C_ISR_NACKF) {
+		dev_dbg(i2c_dev->dev, "<%s>: Receive NACK\n", __func__);
+		writel_relaxed(STM32F7_I2C_ICR_NACKCF, base + STM32F7_I2C_ICR);
+		f7_msg->result = -EBADE;

-ENXIO (see Documentation/i2c/fault-codes)

...

+	timeout = wait_for_completion_timeout(&i2c_dev->complete,
+					      i2c_dev->adap.timeout);
+	ret = f7_msg->result;
+
+	if (!timeout) {
+		dev_dbg(i2c_dev->dev, "Access to slave 0x%x timed out\n",
+			i2c_dev->msg->addr);
+		ret = -ETIMEDOUT;
+	}

Could you rename the variable to time_left? It looks strange, basically:

	if (!timeout)
		return -ETIMEDOUT

...

+	adap->retries = 0;

Why no retries when you check for arbitration lost?

+	adap->algo = &stm32f7_i2c_algo;
+	adap->dev.parent = &pdev->dev;
+	adap->dev.of_node = pdev->dev.of_node;
+
+	init_completion(&i2c_dev->complete);
+
+	ret = i2c_add_adapter(adap);
+	if (ret) {
+		dev_err(&pdev->dev, "Failed to add adapter\n");

Please remove, the core will print info when adding fails.


Rest looks good!

Thanks,

   Wolfram