Re: [PATCH v3 02/11] clk: tegra: retrieve regulator info from framework

From: Jon Hunter
Date: Thu Mar 08 2018 - 17:26:53 EST



On 06/02/18 16:34, Peter De Schrijver wrote:
> The CVB table contains calibration data for the CPU DFLL based on
> process charaterization. The regulator step and offset parameters depend
> on the regulator supplying vdd-cpu , not on the specific Tegra SKU.
> Hence than hardcoding those regulator parameters in the CVB table,
> retrieve them from the regulator framework and store them as part of the
> tegra_dfll_soc_data struct.
>
> Signed-off-by: Peter De Schrijver <pdeschrijver@xxxxxxxxxx>
> ---
> drivers/clk/tegra/clk-dfll.h | 2 ++
> drivers/clk/tegra/clk-tegra124-dfll-fcpu.c | 42 +++++++++++++++++++++++++-----
> drivers/clk/tegra/cvb.c | 16 +++++++++---
> drivers/clk/tegra/cvb.h | 6 ++---
> 4 files changed, 53 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/clk/tegra/clk-dfll.h b/drivers/clk/tegra/clk-dfll.h
> index 83352c8..e7cbc5b 100644
> --- a/drivers/clk/tegra/clk-dfll.h
> +++ b/drivers/clk/tegra/clk-dfll.h
> @@ -21,6 +21,7 @@
> #include <linux/platform_device.h>
> #include <linux/reset.h>
> #include <linux/types.h>
> +#include "cvb.h"
>
> /**
> * struct tegra_dfll_soc_data - SoC-specific hooks/integration for the DFLL driver
> @@ -35,6 +36,7 @@ struct tegra_dfll_soc_data {
> struct device *dev;
> unsigned long max_freq;
> const struct cvb_table *cvb;
> + struct rail_alignment alignment;
>
> void (*init_clock_trimmers)(void);
> void (*set_clock_trimmers_high)(void);
> diff --git a/drivers/clk/tegra/clk-tegra124-dfll-fcpu.c b/drivers/clk/tegra/clk-tegra124-dfll-fcpu.c
> index 269d359..e2dbb79 100644
> --- a/drivers/clk/tegra/clk-tegra124-dfll-fcpu.c
> +++ b/drivers/clk/tegra/clk-tegra124-dfll-fcpu.c
> @@ -22,6 +22,7 @@
> #include <linux/kernel.h>
> #include <linux/init.h>
> #include <linux/platform_device.h>
> +#include <linux/regulator/consumer.h>
> #include <soc/tegra/fuse.h>
>
> #include "clk.h"
> @@ -42,9 +43,6 @@
> .process_id = -1,
> .min_millivolts = 900,
> .max_millivolts = 1260,
> - .alignment = {
> - .step_uv = 10000, /* 10mV */
> - },
> .speedo_scale = 100,
> .voltage_scale = 1000,
> .entries = {
> @@ -82,6 +80,34 @@
> },
> };
>
> +static int get_alignment_from_regulator(struct device *dev,
> + struct rail_alignment *align)
> +{
> + int min_uV, max_uV, n_voltages, ret;
> + struct regulator *reg;
> +
> + reg = devm_regulator_get(dev, "vdd-cpu");
> + if (IS_ERR(reg))
> + return PTR_ERR(reg);
> +
> + ret = regulator_get_constraint_voltages(reg, &min_uV, &max_uV);
> + if (!ret)
> + align->offset_uv = min_uV;
> + else
> + return ret;

Nit-pick ... looks a bit odd, why not ...

if (ret)
return ret;

align->offset_uv = min_uV;

> +
> + align->step_uv = regulator_get_linear_step(reg);
> + if (!align->step_uv && !ret) {

Do you need to test 'ret' here?

> + n_voltages = regulator_count_voltages(reg);
> + if (n_voltages > 1)
> + align->step_uv = (max_uV - min_uV) / (n_voltages - 1);

Later in the patch !align->step_uv is treated as an error, so if
n_voltages should always be greater 1 then why not return an error here?
Seems that this should not happen?

Cheers
Jon

--
nvpublic