Hi Andy,
On 19/09/24 08:27, Andi Shyti wrote:
Hi Chris,
On Wed, Sep 18, 2024 at 11:29:29AM GMT, Chris Packham wrote:
Add support for the I2C controller on the RTL9300 SoC. This is based on
the openwrt implementation[1] but cleaned up to make use of the regmap
APIs.
Can you please add a few more words to describe the device?
Sure will do.
[1] - https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=target/linux/realtek/files-5.15/drivers/i2c/busses/i2c-rtl9300.c
Signed-off-by: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx>
...
+#define I2C_MST_CTRL1 0x0
+#define MEM_ADDR_OFS 8
+#define MEM_ADDR_MASK 0xffffff
+#define SDA_OUT_SEL_OFS 4
+#define SDA_OUT_SEL_MASK 0x7
+#define GPIO_SCL_SEL BIT(3)
+#define RWOP BIT(2)
+#define I2C_FAIL BIT(1)
+#define I2C_TRIG BIT(0)
+#define I2C_MST_CTRL2 0x4
+#define RD_MODE BIT(15)
+#define DEV_ADDR_OFS 8
+#define DEV_ADDR_MASK 0x7f
+#define DATA_WIDTH_OFS 4
+#define DATA_WIDTH_MASK 0xf
+#define MEM_ADDR_WIDTH_OFS 2
+#define MEM_ADDR_WIDTH_MASK 0x3
can we have these masked already shifted? You could use
GENMASK().
I'll take a look.
+#define SCL_FREQ_OFS 0
+#define SCL_FREQ_MASK 0x3
+#define I2C_MST_DATA_WORD0 0x8
+#define I2C_MST_DATA_WORD1 0xc
+#define I2C_MST_DATA_WORD2 0x10
+#define I2C_MST_DATA_WORD3 0x14
Can we use a prefix for all these defines?
Yes will add "RTL9300_".
I assume for the bit values too? So something like "MEM_ADDR_OFS"
becomes "RTL9300_I2C_MST_CTRL1_MEM_ADDR_OFS" is that OK or too verbose?
+
+#define RTL9300_I2C_STD_FREQ 0
+#define RTL9300_I2C_FAST_FREQ 1
This can also be an enum.
Ack
+
+DEFINE_MUTEX(i2c_lock);
...
+static int rtl9300_i2c_write(struct rtl9300_i2c *i2c, u8 *buf, int len)
+{
+ u32 vals[4] = {};
+ int i, ret;
+
+ if (len > 16)
+ return -EIO;
+
+ for (i = 0; i < len; i++) {
+ if (i % 4 == 0)
+ vals[i/4] = 0;
+ vals[i/4] <<= 8;
+ vals[i/4] |= buf[i];
+ }
+
+ ret = regmap_bulk_write(i2c->regmap, i2c->i2c_mst_ofs + I2C_MST_DATA_WORD0,
+ vals, ARRAY_SIZE(vals));
+ if (ret)
+ return ret;
+
+ return len;
why returning "len"? And in any case this is ignored.
I copied that behaviour from the openwrt driver. I think making it the
same as the other functions would make more sense.
+}
+
+static int rtl9300_i2c_writel(struct rtl9300_i2c *i2c, u32 data)
+{
+ int ret;
+
+ ret = regmap_write(i2c->regmap, i2c->i2c_mst_ofs + I2C_MST_DATA_WORD0, data);
+ if (ret)
+ return ret;
+
+ return 0;
return regmap_write(...) ?
In any case, the returned value of these functions is completely
ignored, not even printed. Should we either:
- condier the return value in the _xfer() functions
or
- make all these functions void?
I suppose it's a bit academic. Under the hood it's mmio so it's not as
if it can really fail (famous last words). That said, this switch chip
can be run in a core disabled mode and you could then in theory be doing
I2C over SPI from an external SoC. If someone were just naively updating
a hardware design to add the external SoC they might neglect to move the
I2C connections. It's also just good practice so I'll propagate the
returns up to the _xfer().
+}
+
+static int rtl9300_i2c_execute_xfer(struct rtl9300_i2c *i2c, char read_write,
+ int size, union i2c_smbus_data *data, int len)
+{
+ u32 val, mask;
+ int ret;
+
+ if (read_write == I2C_SMBUS_READ)
+ val = 0;
+ else
+ val = RWOP;
+ mask = RWOP;
+
+ val |= I2C_TRIG;
+ mask |= I2C_TRIG;
how about "mask = RWOP | I2C_TRIG" to make it in one line?
Also val can be simplified as:
val = I2C_TRIG;
if (read_write == I2C_SMBUS_WRITE)
val |= RWOP;
Not a binding commeent, as you wish.
I'll take a look. I kind of did like the pairing of val and mask for
each thing being set.
+
+ ret = regmap_update_bits(i2c->regmap, i2c->i2c_mst_ofs + I2C_MST_CTRL1, mask, val);
+ if (ret)
+ return ret;
+
+ ret = regmap_read_poll_timeout(i2c->regmap, i2c->i2c_mst_ofs + I2C_MST_CTRL1,
+ val, !(val & I2C_TRIG), 100, 2000);
+ if (ret)
+ return ret;
+
+ if (val & I2C_FAIL)
where is val taking taking this bit?
In the regmap_read_poll_timeout().
+ return -EIO;
+
...
+ switch (size) {
+ case I2C_SMBUS_QUICK:
...
+ case I2C_SMBUS_BYTE:
...
+ case I2C_SMBUS_BYTE_DATA:
...
+ case I2C_SMBUS_WORD_DATA:
...
+ case I2C_SMBUS_BLOCK_DATA:
...
+ default:
+ dev_warn(&adap->dev, "Unsupported transaction %d\n", size);
dev_err() ?
Ack.
+ ret = -EOPNOTSUPP;
+ goto out_unlock;
+ }
...
+ switch (clock_freq) {
+ case I2C_MAX_STANDARD_MODE_FREQ:
...
+ case I2C_MAX_FAST_MODE_FREQ:
...
+ default:
+ dev_warn(i2c->dev, "clock-frequency %d not supported\n", clock_freq);
+ return -EINVAL;
If we are returning an error we should print an error, let's make
it a "return dev_err_probe()"
But, I was thinking that by default we can assign
I2C_MAX_STANDARD_MODE_FREQ and if the DTS defines a different
frequency we could just print an error and stick to the default
value. Makes sense?
I don't have a strong opinion. Failing the probe just because something
in the dts is wrong where we can have a sane default does seem overly
harsh. On the other hand I've had hardware QA folks complain when the
I2C bus is running at 98khz instead of 100khz.
+ }
...
+ return i2c_add_adapter(adap);
return devm_i2c_add_adapter(adap);
and the remove function is not needed.
OK thanks. I did look for a devm variant but obviously not hard enough.
+}
+
+static void rtl9300_i2c_remove(struct platform_device *pdev)
+{
+ struct rtl9300_i2c *i2c = platform_get_drvdata(pdev);
+
+ i2c_del_adapter(&i2c->adap);
+}
+
+static const struct of_device_id i2c_rtl9300_dt_ids[] = {
+ { .compatible = "realtek,rtl9300-i2c" },
+ {}
+};
+MODULE_DEVICE_TABLE(of, i2c_rtl9300_dt_ids);
+
+static struct platform_driver rtl9300_i2c_driver = {
+ .probe = rtl9300_i2c_probe,
+ .remove = rtl9300_i2c_remove,
+ .driver = {
+ .name = "i2c-rtl9300",
+ .of_match_table = i2c_rtl9300_dt_ids,
+ },
+};
+
+module_platform_driver(rtl9300_i2c_driver);
+
+MODULE_DESCRIPTION("RTL9300 I2C controller driver");
+MODULE_LICENSE("GPL");
+
Just a trailing blank line here.
Ack.
Thanks,
Andi