Re: [PATCH v4 15/15] iio: adc: xilinx-xadc: Add I2C interface support

["Nagavarapu, Dileep Kumar" <dnagavar@amd.com>]

On 10/19/2026 5:05 AM, Jonathan Cameron wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
> 
> 
> On Mon, 5 Oct 2026 15:12:05 +0530
> Dileep Kumar Nagavarapu <dileepkumar.nagavarapu@amd.com> wrote:
> 
>> Add I2C interface support for Xilinx System Management Wizard IP along
>> with the existing AXI memory-mapped interface. This support enables
>> monitoring the voltage and temperature on UltraScale+ devices where the
>> System Management Wizard is connected via I2C.
>>
>> Key changes:
>> - Implement 32-bit DRP (Dynamic Reconfiguration Port) packet format
>>    as per Xilinx PG185 specification.
>> - Add separate I2C probe with xadc_i2c_of_match_table to handle same
>>    compatible string ("xlnx,system-management-wiz-1.3") on I2C bus.
>> - Add NULL checks for get_dclk_rate callback function in sampling rate
>>    functions to support interfaces without clock control
>> - Create separate iio_info structure(xadc_i2c_info) without event
>>    callbacks for I2C devices
>> - Add xadc_i2c_transaction() function to handle I2C read/write operations
>> - Add XADC_TYPE_US_I2C type to distinguish I2C interface from AXI
>> - Defer XADC hardware initialization until first device access, since the
>>    I2C path reaches the PL SysMon indirectly through a platform controller
>>    DRP bridge that may not yet be operational at probe() time. Initializing
>>    on first read/write instead avoids probe failures caused by transient
>>    unavailability of that downstream path.
>>
>> Co-developed-by: Sai Krishna Potthuri <sai.krishna.potthuri@amd.com>
>> Signed-off-by: Sai Krishna Potthuri <sai.krishna.potthuri@amd.com>
>> Signed-off-by: Dileep Kumar Nagavarapu <dileepkumar.nagavarapu@amd.com>
> Hi Dileep,
> 
> Various comments inline - mostly stuff where we can make things
> a little more readable or extensible.
> 
> Jonathan
> 
>> diff --git a/drivers/iio/adc/xilinx-xadc-core.c b/drivers/iio/adc/xilinx-xadc-core.c
>> index 2addb61cf5b1..80199c4ddf69 100644
>> --- a/drivers/iio/adc/xilinx-xadc-core.c
>> +++ b/drivers/iio/adc/xilinx-xadc-core.c
>> @@ -227,7 +227,7 @@ static int xadc_power_adc_b(struct xadc *xadc, unsigned int seq_mode)
>>         * non-existing ADC-B powers down the main ADC, so just return and don't
>>         * do anything.
>>         */
>> -     if (xadc->ops->type == XADC_TYPE_US)
>> +     if (xadc->ops->type == XADC_TYPE_US || xadc->ops->type == XADC_TYPE_US_I2C)
>>                return 0;
> Maybe add what ever this is similar to below. If it's in the ops structure
> we avoid having to add new types like this.
> 
>>
>>        /* Powerdown the ADC-B when it is not needed. */
>> @@ -249,7 +249,7 @@ static int xadc_get_seq_mode(struct xadc *xadc, unsigned long scan_mode)
>>        unsigned int aux_scan_mode = scan_mode >> 16;
>>
>>        /* UltraScale has only one ADC and supports only continuous mode */
>> -     if (xadc->ops->type == XADC_TYPE_US)
>> +     if (xadc->ops->type == XADC_TYPE_US || xadc->ops->type == XADC_TYPE_US_I2C)
>>                return XADC_CONF1_SEQ_CONTINUOUS;
> 
> Perhaps add this to the ops structure as well as a field called something like
> bool continuous_only;
> 
>>
>>        if (xadc->external_mux_mode == XADC_EXTERNAL_MUX_DUAL)
> 
>>
>>   struct iio_dev *xadc_device_setup(struct device *dev, int size, const struct xadc_ops **ops)
>> @@ -784,7 +797,11 @@ struct iio_dev *xadc_device_setup(struct device *dev, int size, const struct xad
>>                return ERR_PTR(-ENOMEM);
>>
>>        indio_dev->name = xadc_type_names[(*ops)->type];
>> -     indio_dev->info = &xadc_info;
>> +     if ((*ops)->type == XADC_TYPE_US_I2C)
> 
> Why not put the info in the ops structure?  Then you can set it
> unconditionally here.  Generally matching a 'type' ends up
> messier than just adding more stuff to the type specific structure.
> 
>> +             indio_dev->info = &xadc_i2c_info;
>> +     else
>> +             indio_dev->info = &xadc_info;
> 
>> diff --git a/drivers/iio/adc/xilinx-xadc-i2c.c b/drivers/iio/adc/xilinx-xadc-i2c.c
>> new file mode 100644
>> index 000000000000..cee2f7b3f9bd
>> --- /dev/null
>> +++ b/drivers/iio/adc/xilinx-xadc-i2c.c
> 
>> +
>> +#include <linux/iio/iio.h>
>> +
>> +#include "xilinx-xadc.h"
>> +
>> +#define XADC_I2C_READ_DATA_SIZE              2
>> +#define XADC_I2C_WRITE_DATA_SIZE     4       /* 32-bit DRP packet */
> Hmm. Put this as a terminating entry in an enum with the DRP elements
> below. The it becomes clear how they are associated with each other.
> 
>> +#define XADC_I2C_INSTR_READ          BIT(2)
>> +#define XADC_I2C_INSTR_WRITE         BIT(3)
>> +
>> +#define XADC_I2C_DRP_DATA0_MASK              GENMASK(7, 0)
>> +#define XADC_I2C_DRP_DATA1_MASK              GENMASK(15, 8)
>> +#define XADC_I2C_DRP_ADDR_MASK               GENMASK(7, 0)
> All of these seem unnecessary. See below
>> +
>> +/* Byte offsets within the 32-bit DRP packet, per Xilinx PG185 */
>> +#define XADC_I2C_DRP_DATA_LOW                0
>> +#define XADC_I2C_DRP_DATA_HIGH               1
>> +#define XADC_I2C_DRP_ADDR            2
>> +#define XADC_I2C_DRP_INSTR           3
> 
>> +
>> +static int xadc_i2c_read_transaction(struct xadc *xadc, unsigned int reg, u16 *val)
>> +{
>> +     struct xadc_i2c *xadc_i2c = container_of(xadc, struct xadc_i2c, xadc);
>> +     u8 write_buffer[XADC_I2C_WRITE_DATA_SIZE] = { 0 };
> 
> Might as well set the values you have below here rather than init to 0 and then
> override.
> 
>> +     struct i2c_client *client = xadc_i2c->client;
>> +     u8 read_buffer[XADC_I2C_READ_DATA_SIZE];
>> +     int ret;
>> +
>> +     write_buffer[XADC_I2C_DRP_ADDR] = FIELD_GET(XADC_I2C_DRP_ADDR_MASK, reg);
> As below. Not seeing much value in this FIELD_PREP as the reg should always fit
> 
> I think?
> 
>> +     write_buffer[XADC_I2C_DRP_INSTR] = XADC_I2C_INSTR_READ;
>> +
>> +     ret = i2c_master_send(client, write_buffer, XADC_I2C_WRITE_DATA_SIZE);
>> +     if (ret < 0)
>> +             return ret;
>> +
>> +     ret = i2c_master_recv(client, read_buffer, XADC_I2C_READ_DATA_SIZE);
>> +     if (ret < 0)
>> +             return ret;
>> +
>> +     *val = FIELD_PREP(XADC_I2C_DRP_DATA0_MASK, read_buffer[XADC_I2C_DRP_DATA_LOW]) |
>> +            FIELD_PREP(XADC_I2C_DRP_DATA1_MASK, read_buffer[XADC_I2C_DRP_DATA_HIGH]);
>          *val = get_unaligned_le16(&read_buffer[XADC_I2C_DRP_DATA_LOW]);
> 
>> +
>> +     return 0;
>> +}
>> +
>> +static int xadc_i2c_write_transaction(struct xadc *xadc, unsigned int reg, u16 val)
>> +{
>> +     struct xadc_i2c *xadc_i2c = container_of(xadc, struct xadc_i2c, xadc);
>> +     struct i2c_client *client = xadc_i2c->client;
>> +     u8 write_buffer[XADC_I2C_WRITE_DATA_SIZE];
>> +     int ret;
>> +
>> +     /* low byte of the 16-bit DRP data value */
>> +     write_buffer[XADC_I2C_DRP_DATA_LOW] = FIELD_GET(XADC_I2C_DRP_DATA0_MASK, val);
>> +     /* high byte of the 16-bit DRP data value */
>> +     write_buffer[XADC_I2C_DRP_DATA_HIGH] = FIELD_GET(XADC_I2C_DRP_DATA1_MASK, val);
> 
>          put_unaligned_le16(val, &write_buffer[XADC_I2C_DRP_DATA_LOW]);
> and no comments needed.
> 
>> +     write_buffer[XADC_I2C_DRP_ADDR] = FIELD_GET(XADC_I2C_DRP_ADDR_MASK, reg);
> 
> Given this is just a clamp to size of something that I assume always fits,
> I'd drop the FIELD_PREP as providing no value.
> 
>> +     write_buffer[XADC_I2C_DRP_INSTR] = XADC_I2C_INSTR_WRITE;
>> +
>> +     ret = i2c_master_send(client, write_buffer, XADC_I2C_WRITE_DATA_SIZE);
> 
> I guess in this particular case we know what the i2c master is so can assume
> this never runs short?  (in practice they never do but I think the interface
> allows for that case).
> 

Yes, I'll add a check for ret != XADC_I2C_WRITE_DATA_SIZE and return 
-EIO for a short write

>> +     if (ret < 0)
>> +             return ret;
>> +
>> +     return 0;
>> +}
> 
> 
> 
>> +
>> +static int xadc_i2c_probe(struct i2c_client *client)
>> +{
>> +     struct device *dev = &client->dev;
>> +     unsigned int conf0, bipolar_mask;
>> +     const struct xadc_ops *ops;
>> +     struct iio_dev *indio_dev;
>> +     struct xadc_i2c *xadc_i2c;
>> +     struct xadc *xadc;
>> +     int ret;
>> +
>> +     indio_dev = xadc_device_setup(dev, sizeof(*xadc_i2c), &ops);
>> +     if (IS_ERR(indio_dev))
>> +             return PTR_ERR(indio_dev);
>> +
>> +     xadc_i2c = iio_priv(indio_dev);
>> +     xadc_i2c->client = client;
>> +     xadc = &xadc_i2c->xadc;
> 
> It might be worth a little helper for here...
> 
>> +     xadc->ops = ops;
>> +
>> +     ret = devm_mutex_init(dev, &xadc->mutex);
>> +     if (ret)
>> +             return ret;
>> +
>> +     spin_lock_init(&xadc->lock);
> 
> to here to initialize the xadc structure as I don't think
> the individual drivers should really know about xadc->lock etc.
> 
> 
>> +
>> +     ret = xadc_device_configure(indio_dev, 0, &conf0, &bipolar_mask);
>> +     if (ret)
>> +             return dev_err_probe(dev, ret, "Failed to setup the device\n");
>> +
>> +     xadc_i2c->conf0 = conf0;
>> +     xadc_i2c->bipolar_mask = bipolar_mask;
>> +     xadc_i2c->hw_initialized = false;
>> +
>> +     return devm_iio_device_register(dev, indio_dev);
>> +}
>> +
>> +static const struct of_device_id xadc_i2c_of_match_table[] = {
>> +     {
>> +             .compatible = "xlnx,system-management-wiz-1.3",
>> +             .data = &xadc_system_mgmt_wiz_i2c_ops,
>> +     },
>> +     {}
> Trivial but local IIO preference is
>          { }
> 
> I wanted to standardise these on something, and happened to pick
> that a few years back.
> 
>> +};
>> +MODULE_DEVICE_TABLE(of, xadc_i2c_of_match_table);
> 

Thanks
Dileep