Re: [PATCH] mfd: Add Cherrytrail WhiskeyCove PMIC driver

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 27-02-17 23:04, Andy Shevchenko wrote:
> On Mon, Feb 27, 2017 at 12:18 PM, Hans de Goede <hdegoede@redhat.com> wrote:
>> Add mfd driver for Intel CHT WhiskeyCove PMIC, based on various non
>> upstreamed CHT WhiskeyCove PMIC patches. For now this just adds a minimal
>> version which implements just enough to get ACPI PMIC opregion support to
>> work, so that suspend/resume will work on machines with this PMIC.
>
> Whiskey Cove in Subject and commit message.

Will fix for v3.

>> +intel-soc-pmic-objs            := intel_soc_pmic_core.o intel_soc_pmic_crc.o intel_soc_pmic_chtwc.o
>>  intel-soc-pmic-$(CONFIG_INTEL_PMC_IPC) += intel_soc_pmic_bxtwc.o
>>  obj-$(CONFIG_INTEL_SOC_PMIC)   += intel-soc-pmic.o
>
> Can we use one module per driver? I already pointed out to my patch
> (unfortunately needs to be updated) that tries to fix it for BXT WC.

Will fix for v3.

>> +/* PMIC device registers */
>
>> +#define REG_ADDR_MASK          0xff00
>
> GENMASK()
>
>> +#define REG_ADDR_SHIFT         8
>> +#define REG_OFFSET_MASK                0xff
>
> Ditto.

Both fixed for v3.

>> +
>> +/* Whiskey Cove PMIC share same ACPI ID between different platforms */
>> +#define CHT_WC_HRV             3
>> +
>> +static struct mfd_cell cht_wc_dev[] = {
>> +       {
>> +               .name = "cht_wcove_region",
>> +       },
>> +};
>> +
>> +/*
>> + * The CHT Whiskey Cove covers multiple i2c addresses, with a 1 byte
>> + * register address space per i2c address, so we use 16 bit register
>> + * addresses where the high 8 bits contain the i2c client address.
>> + */
>> +static int cht_wc_byte_reg_read(void *context, unsigned int reg,
>> +                               unsigned int *val)
>> +{
>> +       struct i2c_client *client = context;
>
>> +       int ret, orig_addr = client->addr;
>> +
>> +       if (reg & REG_ADDR_MASK)
>> +               client->addr = (reg & REG_ADDR_MASK) >> REG_ADDR_SHIFT;
>> +       else
>> +               client->addr = CHT_WC_DEVICE1_ADDR;
>> +       ret = i2c_smbus_read_byte_data(client, reg & REG_OFFSET_MASK);
>> +       client->addr = orig_addr;
>
> Looks a bit hackish to me.

Agreed.

> Why not to define DEVICE1_ADDR as 0x6e00, for example?

This is intended for callers who specify a register offset
only without specifying which Whiskey Cove sub-device
(i2c address) they want. This comes from the out of tree code
I based this on, where this is also actually (ab)used by some of
the mfd child device drivers. I plan for all the mainlined
child device drivers to properly use a fill 16 bit address,
which only leaves the ACPI REGS opregion accesses as potentially
leaving the upper 8 bits 0. All the DSDTs I've access to do
always properly use a 16 bit address. So I will just drop this
hack for v3 replacing it with:

if (!(reg & REG_ADDR_MASK)) {
	dev_err(&client->dev, "i2c device address not specified\n");
	return -EINVAL;
}

Which will catch and invalid REGS opregion accesses.




>
>> +
>> +       if (ret < 0)
>> +               return ret;
>> +
>> +       *val = ret;
>> +       return 0;
>> +}
>> +
>> +static int cht_wc_byte_reg_write(void *context, unsigned int reg,
>> +                                unsigned int val)
>> +{
>> +       struct i2c_client *client = context;
>> +       int ret, orig_addr = client->addr;
>> +
>> +       if (reg & REG_ADDR_MASK)
>> +               client->addr = (reg & REG_ADDR_MASK) >> REG_ADDR_SHIFT;
>> +       else
>> +               client->addr = CHT_WC_DEVICE1_ADDR;
>> +       ret = i2c_smbus_write_byte_data(client, reg & REG_OFFSET_MASK, val);
>> +       client->addr = orig_addr;
>
> Ditto.
>
>> +
>> +       return ret;
>> +}
>
>> +static int cht_wc_probe(struct i2c_client *client,
>> +                       const struct i2c_device_id *i2c_id)
>> +{
>> +       struct device *dev = &client->dev;
>> +       struct intel_soc_pmic *pmic;
>> +       acpi_handle handle;
>> +       acpi_status status;
>> +       unsigned long long hrv;
>> +       int ret;
>> +
>
>> +       handle = ACPI_HANDLE(dev);
>
> Useless temporary variable?

Will fix for v3.

>
>> +       status = acpi_evaluate_integer(handle, "_HRV", NULL, &hrv);
>> +       if (ACPI_FAILURE(status)) {
>> +               dev_err(dev, "Failed to get PMIC hardware revision\n");
>> +               return -ENODEV;
>> +       }
>> +       if (hrv != CHT_WC_HRV) {
>> +               dev_err(dev, "Invalid PMIC hardware revision: %llu\n", hrv);
>> +               return -ENODEV;
>> +       }
>> +       if (client->irq < 0) {
>> +               dev_err(dev, "Invalid IRQ\n");
>> +               return -ENODEV;
>> +       }
>> +
>
>> +       pmic->regmap = devm_regmap_init(dev, NULL, client, &cht_wc_regmap_cfg);
>> +       if (IS_ERR(pmic->regmap)) {
>> +               ret = PTR_ERR(pmic->regmap);
>
>> +               dev_err(dev, "Failed to initialise regmap: %d\n", ret);
>
> Is it anyhow useful?

Nope, I will remove this for v3.

>
>> +               return ret;
>> +       }
>> +
>
>> +       ret = mfd_add_devices(dev, -1, cht_wc_dev, ARRAY_SIZE(cht_wc_dev),
>> +                             NULL, 0, NULL);
>> +       if (ret) {
>> +               dev_err(dev, "Failed to add devices: %d\n", ret);
>> +               return ret;
>> +       }
>> +
>> +       return 0;
>
> return devm_mfd_add_devices(...);

Will fix for v3.

>> +}
>
>> +++ b/include/linux/mfd/intel_chtwc.h
>> @@ -0,0 +1,75 @@
>> +/*
>
>> + * intel_chtwc.h - Header file for Intel Cherrytrail Whiskey Cove PMIC
>
> Remove file name.

Will fix for v3.

Regards,

Hans