On 6/24/24 09:13, Farouk Bouabid wrote:
Hi Guenter,
On 19.06.24 15:31, Guenter Roeck wrote:
On 6/19/24 00:45, Farouk Bouabid wrote:
If it is properly defined in devicetree, the emulated AMC6821 should be
an i2c device, possibly sitting behind an i2c multiplexer, not a
platform device.
The emulated AMC6821 and the Mule I2C mux are both reachable using I2C address (0x18), and hence the use of MFD as the mux only uses one I2C register that is not used by AMC6821.
Whatever you do, the amc chip is still an i2c driver and needs to remain one.
Modeling it as platform driver is simply wrong, and I won't accept those patches.
The issue that we have cannot be handled by an I2C mux because in that case both the mux and its child would have the same address which is not supported in the I2C subsystem:
i2c-mux@18 {
compatible = "tsd,mule-i2c-mux";
reg = <0x18>;
#address-cells = <1>;
#size-cells = <0>;
i2c10: i2c@0 {
reg = <0x0>;
#address-cells = <1>;
#size-cells = <0>;
fan: fan@18 {
compatible = "ti, amc6821";
reg = <0x18>;
};
};
};
The I2C maintainer rejected supporting this use case and suggested that an MFD could probably be more suitable.
On one hand, the MFD looks indeed more appropriate and a lot of I2C devices are modeled through platform sub devices. On the other hand we are emulating the amc6821 in our device which requires us to have it modeled as platform:
The difference is that those other i2c devices are real multi-function
devices.
+--------+----------------+------------------------------+
| Mule (MFD) |
0x18 | +----------------+ |
--------+----->| amc6821 | |
| | +----------------+ |
| +----->| Mux |-----+ |
| +----------------+ | |
| V__ +---------+ |
| | \-------->| isl1208 | |
| | | +---------+ |
0x6f | | M |-------->| dev #1 | |
------------------------------------>| U | +---------+ |
| | X |-------->| dev #2 | |
| | | +---------+ |
| | /-------->| dev #3 | |
| |__/ +---------+ |
+--------------------------------------------------------+
It would have been much more appropriate to use a different I2C address for the mux.
If we cannot proceed with that then we could add a compatible to the amc6821 driver to add the mux device (Basically the "tsd,mule" compatible in amc6821 compatible list would be a combo driver with mux logic + amc6821). Do you think that is more appropriate ?
Implement the mux as part of the amc6821 driver ? No. We could discuss
instantiating the i2c mux driver from the amc6821 driver.
Guenter
