Re: [PATCH v2 2/3] hwmon (xdpe12284): Add support for xdpe11280

[sylv <sylv@sylv.io>]

On Thu, 2022-02-17 at 10:11 -0800, Guenter Roeck wrote:
> On 2/17/22 08:39, sylv wrote:
> > On Thu, 2022-02-17 at 07:20 -0800, Guenter Roeck wrote:
> > > On 2/17/22 06:41, Marcello Sylvester Bauer wrote:
> > > > Add support for another family of Infineon Multi-phase
> > > > controllers.
> > > > The xdpe11280 uses linear instead of vid data format for VOUT.
> > > > Add the required logic to detect VOUT_MODE during
> > > > identification,
> > > > and set the format accordingly.
> > > > 
> > > This talks about the chip, but ...
> > 
> > Oh, sure. I should be more clear that it is a chip of a different
> > family.
> > 
> > > 
> > > > Signed-off-by: Marcello Sylvester Bauer <sylv@sylv.io>
> > > > ---
> > > >    Documentation/hwmon/xdpe12284.rst | 10 +++++++---
> > > >    drivers/hwmon/pmbus/xdpe12284.c   | 17 ++++++++++++++++-
> > > >    2 files changed, 23 insertions(+), 4 deletions(-)
> > > > 
> > > > diff --git a/Documentation/hwmon/xdpe12284.rst
> > > > b/Documentation/hwmon/xdpe12284.rst
> > > > index 67d1f87808e5..e9a5a19b0c3d 100644
> > > > --- a/Documentation/hwmon/xdpe12284.rst
> > > > +++ b/Documentation/hwmon/xdpe12284.rst
> > > > @@ -5,6 +5,10 @@ Kernel driver xdpe122
> > > >    
> > > >    Supported chips:
> > > >    
> > > > +  * Infineon XDPE11280
> > > > +
> > > > +    Prefix: 'xdpe11280'
> > > > +
> > > >      * Infineon XDPE12254
> > > >    
> > > >        Prefix: 'xdpe12254'
> > > > @@ -20,9 +24,9 @@ Authors:
> > > >    Description
> > > >    -----------
> > > >    
> > > > -This driver implements support for Infineon Multi-phase
> > > > XDPE122
> > > > family
> > > > -dual loop voltage regulators.
> > > > -The family includes XDPE12284 and XDPE12254 devices.
> > > > +This driver implements support for Infineon Multi-phase
> > > > XDPE112
> > > > and XDPE122
> > > > +family dual loop voltage regulators.
> > > > +These families include XDPE11280, XDPE12284 and XDPE12254
> > > > devices.
> > > >    The devices from this family complaint with:
> > > 
> > > Hmm, s/complaint/compliant/.
> > 
> > ack
> > 
> > > 
> > > Anyway, is it still correct that the XDPE11280 is compliant with
> > > the
> > > Intel specifications ? The datasheet isn't public, so I can not
> > > check
> > > myself
> > > what the chip actually supports.
> > 
> > The hardware we are testing on is an Intel Cascade Lake, which uses
> > this chip. The datasheet tells us it is compliant:
> > 
> > ```
> > The XDPE11280B implements a fully compliant Intel® VR 13, VR12, VR
> > 12.5, VR13.HC, and IMVP8 Serial VID (SVID)
> > interface.
> > ```
> > 
> 
> This seems to be one of those chips where the vendor doesn't even
> admit
> that it exists unless one signs an NDA. Sigh :-(.

Yeah, it's pretty... We do not understand the reasoning.

> 
> > > 
> > > >    
> > > >    - Intel VR13 and VR13HC rev 1.3, IMVP8 rev 1.2 and IMPVP9
> > > > rev 1.3
> > > > DC-DC
> > > > diff --git a/drivers/hwmon/pmbus/xdpe12284.c
> > > > b/drivers/hwmon/pmbus/xdpe12284.c
> > > > index b07da06a40c9..ec947c697670 100644
> > > > --- a/drivers/hwmon/pmbus/xdpe12284.c
> > > > +++ b/drivers/hwmon/pmbus/xdpe12284.c
> > > > @@ -27,6 +27,9 @@ static int xdpe122_read_word_data(struct
> > > > i2c_client *client, int page,
> > > >          s32 mantissa;
> > > >          int ret;
> > > >    
> > > > +       if (info->format[PSC_VOLTAGE_OUT] == linear)
> > > > +               return -ENODATA;
> > > > +
> > > >          switch (reg) {
> > > >          case PMBUS_VOUT_OV_FAULT_LIMIT:
> > > >          case PMBUS_VOUT_UV_FAULT_LIMIT:
> > > > @@ -75,9 +78,19 @@ static int xdpe122_read_word_data(struct
> > > > i2c_client *client, int page,
> > > >    static int xdpe122_identify(struct i2c_client *client,
> > > >                              struct pmbus_driver_info *info)
> > > >    {
> > > > -       u8 vout_params;
> > > > +       u8 vout_mode, vout_params;
> > > >          int i, ret;
> > > >    
> > > > +       ret = pmbus_read_byte_data(client, 0, PMBUS_VOUT_MODE);
> > > > +       if (ret < 0)
> > > > +               return ret;
> > > > +
> > > > +       vout_mode = ret >> 5;
> > > > +       if (vout_mode == 0) {
> > > > +               info->format[PSC_VOLTAGE_OUT] = linear;
> > > > +               return 0;
> > > > +       }
> > > 
> > > .... the code above is not chip dependent but this code checks
> > > the
> > > mode register on page 0. At the very least this is confusing.
> > > 
> > > What the driver should do instead is to check the chip type
> > > during
> > > probe and make the necessary changes there, ie set the output
> > > format
> > > for xdpe11280 to linear and drop the pointers to the .identify
> > > and .read_word_data.
> > 
> > Sgtm. This sounds like a better approach indeed.
> > 
> 
> That makes me wonder if the chip needs to be added to this driver in
> the first
> place, or if it could be added to pmbus.c instead. Any idea ?

Oh, we did wrote a standalone driver too, and it works fine.
Maybe it's better to upsteam it instead. :)

> 
> Thanks,
> Guenter
> 
> > Thanks,
> > Marcello
> > 
> > > 
> > > Guenter
> > > 
> > > > +
> > > >          for (i = 0; i < XDPE122_PAGE_NUM; i++) {
> > > >                  /* Read the register with VOUT scaling
> > > > value.*/
> > > >                  ret = pmbus_read_byte_data(client, i,
> > > > PMBUS_VOUT_MODE);
> > > > @@ -140,6 +153,7 @@ static int xdpe122_probe(struct i2c_client
> > > > *client)
> > > >    }
> > > >    
> > > >    static const struct i2c_device_id xdpe122_id[] = {
> > > > +       {"xdpe11280", 0},
> > > >          {"xdpe12254", 0},
> > > >          {"xdpe12284", 0},
> > > >          {}
> > > > @@ -148,6 +162,7 @@ static const struct i2c_device_id
> > > > xdpe122_id[]
> > > > = {
> > > >    MODULE_DEVICE_TABLE(i2c, xdpe122_id);
> > > >    
> > > >    static const struct of_device_id __maybe_unused
> > > > xdpe122_of_match[] = {
> > > > +       {.compatible = "infineon,xdpe11280"},
> > > >          {.compatible = "infineon,xdpe12254"},
> > > >          {.compatible = "infineon,xdpe12284"},
> > > >          {}
> > > 
> > 
>