i3c: scanning for i2c client devices on the network

[Boris Staletic]

On Tue, Sep 16, 2025 at 01:08:44PM -0400, Frank Li wrote:
> On Wed, Sep 10, 2025 at 02:01:33PM +0000, Boris Staletic wrote:
> > Hello,
> >
> > We have an I3C bus where the user may ultimately attach an I2C slave with an arbitrary address, with the idea
> > of using i2c-utils to communicate with the I2C slave from userland.
> >
> > With the current kernel, I2C slaves need to be known statically and thus i2cdetect fails to detect any slaves.
> > All communication attempts get rejected early, because i3c_master_find_i2c_dev_by_addr returns NULL and then i3c_master_i2c_adapter_xfer exits with ENOENT.
> > I3c_master_find_i2c_dev_by_addr returns NULL because bus->devs.i2c ends up being empty.
> >
> > Since the I2C drivers have i2c_detect and i3c_master_controller has its own i2c_adater, I looked into what needs to be done for networks
> > with I3C masters to use the existing i2c_detect mechanism. That lead me to the following list of changes:
> >
> > - i3c_master_controllr's i2c_adapter needs to have its class member set.
> > - In i2c_detect_address, temp_client->dev needs to be populated and registered before the call to i2c_default_probe.
> >   - Otherwise the attempt to probe gets rejected early, for the same reason described above (i3c_master_find_i2c_dev_by_addr returning NULL).
> >
> > That much was enough to get i2c_detect to work with an I3C master.
> > To complete the use case I also needed a registered i2c_driver that implements a simple i2c_driver.detect function.
> > In my test, the detect function simply checked whether i2c_smbus_read_byte_data(client, 0) returns a non-error value.
> >
> > Could anyone comment whether this is the right approach for implementing I2C slave detection with an I3C master?
> 
> Generally, it is the same as I2C. just send out address, and check if target
> ACK/NACK this address.

Are you referring here to i2c_new_scanned_device? If so, that also, in
my testing, couldn't detect any devices without modificatioins.
The failure to scan happens because:

- i2c_new_scanned_device calls i2c_default_probe, which calls
  i2c_smbus_xfer.
- that eventually gets to i3c_master_i2c_adapter_xfer, which looks up
  existing devices by calling i3c_master_find_i2c_dev_by_addr.
- i3c_master_find_i2c_dev_by_addr returns NULL
- i3c_master_i2c_adapter_xfer then returns -ENOENT, ven before trying to
  communicate over the wire.

I am probably missing something, because, as far as I can tell, neither
i2c_detect, nor i2c_new_scanned_device can currently be used with an I3C
master.

Would you mind elaborating more on the suggested approach?

Thanks in advance,
Boris Staletic

> 
> Address 7E, REPEAT START, scan's address. ACK/NACK STOP.
> 
> Check ACK/NACK. But you don't know if it is i3c device or i2c device by this
> way.
> 
> send 7E to avoid IBI during you send out address.
> 
> Fank
> 
> > The most iffy part of this approach is the need for such a dummy driver, just so the I3C master knows that "something exists".
> > I'd welcome any sort of feedback.
> >
> > Thanks,
> > Boris Staletic
> > --
> > linux-i3c mailing list
> > linux-i3c at lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/linux-i3c