On Wed, 16 Aug 2023 16:02:25 +0800
Mingjin Yang <mingjin.yang@xxxxxxxxxx> wrote:
> This patch adds support for UNISOC UMP family
> of ADC driver. The ADC peripheral can support both
> voltage and current mode whose input signal is
> connected to the PMIC ADC AMUX.
>
> Signed-off-by: Mingjin Yang <mingjin.yang@xxxxxxxxxx>
Hi Mingjin,
Thanks for your driver - it's in pretty good shape for a v0.
Comments inline.
Jonathan
> diff --git a/drivers/iio/adc/sprd_pmic_adc.c b/drivers/iio/adc/sprd_pmic_adc.c
> new file mode 100644
> index 000000000000..6113bd84cd77
> --- /dev/null
> +++ b/drivers/iio/adc/sprd_pmic_adc.c
> @@ -0,0 +1,1298 @@
> +// SPDX-License-Identifier: GPL-2.0
> +// Copyright (C) 2020 Unisoc Inc.
> +
> +#include <linux/hwspinlock.h>
> +#include <linux/iio/iio.h>
> +#include <linux/math64.h>
> +#include <linux/mfd/syscon.h>
> +#include <linux/module.h>
> +#include <linux/nvmem-consumer.h>
> +#include <linux/of.h>
> +#include <linux/of_device.h>
> +#include <linux/platform_device.h>
> +#include <linux/regmap.h>
> +#include <linux/regulator/consumer.h>
> +#include <linux/slab.h>
> +#include <linux/sort.h>
> +
> +/* ADC controller registers definition */
> +#define SPRD_ADC_CTL 0x0
> +
> +#define SPRD_ADC_CH_CFG 0x4
> +#define SPRD_ADC_CHN_ID_MASK GENMASK(4, 0)
> +
> +#define SPRD_ADC_DATA 0x4c
> +#define SPRD_ADC_DATA_MASK GENMASK(11, 0)
> +
> +#define SPRD_ADC_INT_EN 0x50
> +#define SPRD_ADC_IRQ_EN BIT(0)
> +
> +#define SPRD_ADC_INT_CLR 0x54
> +#define SPRD_ADC_IRQ_CLR BIT(0)
> +
> +#define SPRD_ADC_INT_STS 0x58
> +
> +#define SPRD_ADC_INT_RAW 0x5c
> +#define SPRD_ADC_IRQ_RAW BIT(0)
> +
> +/* Bits and mask definition for SPRD_ADC_CTL register */
> +#define SPRD_ADC_EN BIT(0)
> +#define SPRD_ADC_CHN_RUN BIT(1)
> +#define SPRD_ADC_12BIT_MODE BIT(2)
> +#define SPRD_ADC_RUN_NUM_MASK GENMASK(7, 4)
> +#define SPRD_ADC_RUN_NUM_SHIFT 4
> +#define SPRD_ADC_AVERAGE_SHIFT 8
> +#define SPRD_ADC_AVERAGE_MASK GENMASK(10, 8)
> +
> +/* Timeout (ms) for the trylock of hardware spinlocks */
> +#define SPRD_ADC_HWLOCK_TIMEOUT 5000
Name it so the units are CLAR
SPRD_ADC_HWLOCK_TIMEOUT_MS
then drop teh comment that tells us nothing.
> +
> +/* Maximum ADC channel number */
Obvious comment: drop it.
> +#define SPRD_ADC_CHANNEL_MAX 32
> +
> +/* Timeout (us) for ADC data conversion according to ADC datasheet */
Reference specific part of datasheet or drop the comment.
> +#define SPRD_ADC_RDY_TIMEOUT 1000000
Name it so the unit is obviously
TIMEOUT_US
> +#define SPRD_ADC_POLL_RAW_STATUS 500
What's this one?
> +
> +/* ADC voltage ratio definition */
> +#define SPRD_RATIO_NUMERATOR_OFFSET 16
> +#define SPRD_RATIO_DENOMINATOR_MASK GENMASK(15, 0)
> +#define RATIO(n, d) (((n) << SPRD_RATIO_NUMERATOR_OFFSET) | (d))
> +#define R_INVALID (0xffffffff)
> +#define RATIO_DEF RATIO(1, 1)
Prefix every define..
> +
> +/* ADC specific channel reference voltage 3.5V */
> +#define SPRD_ADC_REFVOL_VDD35 3500000
> +
> +/* ADC default channel reference voltage is 2.8V */
> +#define SPRD_ADC_REFVOL_VDD28 2800000
> +
> +#define SPRD_ADC_CELL_MAX (2)
> +#define SPRD_ADC_INVALID_DATA (0XFFFFFFFF)
> +#define SPRD_ADC_INIT_MAGIC (0xa7a77a7a)
> +#define ADC_MESURE_NUMBER_SW (15)
> +#define ADC_MESURE_NUMBER_HW_DEF (3) /* 2 << 3 = 8 times */
Prefix with SPRD and also the brackets around integer values have little purpose
so drop them. Note the ones around variables are good.
> +
> +#define CH_DATA_FMT "ch%d_d[(%d %d 0x%x 0x%x) | (%u/%u), (%u/%u), (%u/%u), (%u/%u), (%u/%u)]\n"
This define hurts readability. Just put in inline in the one place it's used for debug.
> +
> +#define SPRD_ADC_CHANNEL(index, mask) { \
> + .type = IIO_VOLTAGE, \
> + .channel = index, \
> + .info_mask_separate = mask | BIT(IIO_CHAN_INFO_SCALE), \
> + .datasheet_name = "CH##index", \
> + .indexed = 1, \
> +}
> +
> +#define CH_DATA_INIT(sl, filter, isen, ip_r) \
> +{ \
> + .scale = sl, \
> + .isen_info = isen, \
> + .filter_info = filter, \
> + .ip_ratio = ip_r, \
> + .ch_aux_ratio = {0, 0, 0, 0}, \
> + .inited = SPRD_ADC_INIT_MAGIC, \
> +}
> +
> +#define CALIB_INFO_INIT(ch, sl, gph) \
> +{ \
> + .channel = ch, \
> + .scale = sl, \
> + .graph = gph, \
> + .inited = SPRD_ADC_INIT_MAGIC, \
> +}
> +
> +#define GRAPH_POINT_INIT(v0, a0, v1, a1)\
space before \
also why the blank line here?
> + \
> + .volt0 = v0, \
> + .adc0 = a0, \
> + .volt1 = v1, \
> + .adc1 = a1, \
> + .inited = SPRD_ADC_INIT_MAGIC
> +
> +
> +enum SPRD_PMIC_TYPE {
> + UMP9620_ADC,
> + UMP518_ADC,
> +};
> +
> +enum SPRD_ADC_GRAPH_TYPE {
> + TWO_CELL_GRAPH,
> + SPRD_ADC_GRAPH_TYPE_MAX
> +};
> +
> +enum SPRD_ADC_GRAPH_OFFSET {
> + GRAPH_BIG,
> + GRAPH_SMALL,
> + GRAPH_VBAT_DET,
> + GRAPH_MAX
> +};
> +
> +enum SPRD_ADC_SCALE {
> + SCALE_00,
> + SCALE_01,
> + SCALE_10,
> + SCALE_11,
> + SPRD_ADC_SCALE_MAX
> +};
> +
> +enum SPRD_ADC_REG_TYPE {
> + REG_MODULE_EN,
> + REG_CLK_EN,
> + REG_SOFT_RST,
> + REG_XTL_CTRL,
> + REG_SCALE,
> + REG_CFG_FIXED_ST = 11,/* FIXED CFG START */
> + REG_CAL_EN,
> + REG_NTC_MUX,
> + REG_CFG_FIXED_END,/* FIXED CFG END */
> + REG_ISEN_ST = 21,/* CURRENT MODE START */
> + REG_ISEN0,
> + REG_ISEN1,
> + REG_ISEN2,
> + REG_ISEN3,
> + REG_ISEN_END,/* CURRENT MODE END */
> + SPRD_ADC_REG_TYPE_MAX
> +};
> +
> +enum SPRD_ADC_REG_BASE {
> + BASE_GLB,
> + BASE_ANA
> +};
> +
> +struct sprd_adc_pm_data {
> + struct regmap *clk_regmap;
> + u32 clk_reg;/* adc clk26 vote reg */
> + u32 clk_reg_mask;/* adc clk26 vote reg mask */
> + bool dev_suspended;
> +};
> +
> +/*bit[0-7]: scale
/*
* bit[0-7]: scale
...
*/
However, what is it referring to?
> + *bit[8-15]: filter_info(bit8: sw filter support, bit[9-15]: hw filter val(2<<n))
> + *bit[16-23]: isen_info (bit16: isen support, bit[17-23]: isen val)
> + */
> +struct sprd_adc_channel_data {
> + int scale;
> + int graph;
> + int isen_info;
> + int filter_info;
> + int ip_ratio;
> + int ch_aux_ratio[SPRD_ADC_SCALE_MAX];
> + int inited;
> +};
> +
> +struct calib_info {
> + int channel;
> + int scale;
> + int graph;
> + int ch_aux_ratio[SPRD_ADC_SCALE_MAX];
> + int inited;
> +};
> +
> +struct reg_bit {
sprd_adc_reg_bit etc
to reduce chance of a clash on naming in future.
> + u32 base;
> + u32 reg_addr;
> + u32 mask;
> + u32 offset;
> + u32 val_set;
> + u32 val_clr;
> + u32 inited;
> + bool reverse;
> + u32 (*get_val)(void *pri, int ch, bool set);
> +};
> +
> +static void sprd_adc_calib_with_two_cell(struct sprd_adc_linear_graph *graph);
> +static u32 get_isen(void *pri, int ch, bool enable);
> +static int sprd_adc_soft_rst(struct sprd_adc_data *data);
> +static int sprd_adc_hw_enable(struct sprd_adc_data *data);
blank line here.
> +static inline u32 GET_REG_ADDR(struct sprd_adc_data *data, int index)
Unless you have a very strong reason don't tell the compiler what to inline.
These hints are often limit what it can optimise somewhat and can hurt
the actual performance.
Generic naming without a prefix tends to break long term. Also, this is
just a function, shouldn't be in capitals.
sprd_adc_get_reg_addr()
> +{
> + u32 base = ((data->var_data->reg_list[index].base == BASE_GLB)
> + ? (data->var_data->glb_reg_base)
> + : (data->base - data->var_data->adc_reg_base_offset));
> + return (base + data->var_data->reg_list[index].reg_addr);
> +}
> +
> +static inline void sprd_adc_ch_data_show(struct sprd_adc_data *data, int ch)
> +{
> + struct sprd_adc_channel_data *ch_data = &data->ch_data[ch];
> + struct u32_fract ip_fract, aux_fract0, aux_fract1, aux_fract2, aux_fract3;
> +
> + ratio_to_u32_fract(ch_data->ip_ratio, &ip_fract);
> + ratio_to_u32_fract(ch_data->ch_aux_ratio[0], &aux_fract0);
> + ratio_to_u32_fract(ch_data->ch_aux_ratio[1], &aux_fract1);
> + ratio_to_u32_fract(ch_data->ch_aux_ratio[2], &aux_fract2);
> + ratio_to_u32_fract(ch_data->ch_aux_ratio[3], &aux_fract3);
> + dev_dbg(data->dev, CH_DATA_FMT,
> + ch, ch_data->scale, ch_data->graph, ch_data->isen_info, ch_data->filter_info,
> + ip_fract.numerator, ip_fract.denominator,
> + aux_fract0.numerator, aux_fract0.denominator,
> + aux_fract1.numerator, aux_fract1.denominator,
> + aux_fract2.numerator, aux_fract2.denominator,
> + aux_fract3.numerator, aux_fract3.denominator);
A lot of code in this driver is debug. Is it actually useful once you've finished
development? I'd consider dropping at least some of it to reduce the amount of
code you need to maintain long term.
> +}
> +
> +static void sprd_adc_ch_data_merge(struct sprd_adc_data *data, struct sprd_adc_channel_data *d_diff,
Long line - generally try to stay under 80 chars if it doesn't hurt readability
and under 100 unless it's a message someone might want to grep for.
> + struct sprd_adc_channel_data *d_def)
> +{
> + struct sprd_adc_channel_data *ch_data;
> + int ch;
> +
> + for (ch = 0; ch < SPRD_ADC_CHANNEL_MAX; ch++) {
> + ch_data = &data->ch_data[ch];
> + *ch_data = ((d_diff[ch].inited == SPRD_ADC_INIT_MAGIC) ? d_diff[ch] : *d_def);
> + }
> +
> + for (ch = 0; ch < SPRD_ADC_CHANNEL_MAX; ch++) {
> + sprd_adc_aux_ratio_init(data, ch);
> + sprd_adc_ch_data_show(data, ch);
> + }
> +}
> +static void ump9620_ch_data_init(struct sprd_adc_data *data)
> +{
> + struct sprd_adc_channel_data ch_data_def = CH_DATA_INIT(SCALE_00, 0, 0, RATIO_DEF);
> + struct sprd_adc_channel_data ch_data[SPRD_ADC_CHANNEL_MAX] = {
const
> + [0] = CH_DATA_INIT(SCALE_01, 0, 0, RATIO_DEF),
> + [5] = CH_DATA_INIT(SCALE_00, 0x1, 0x9, RATIO_DEF),
> + [6] = CH_DATA_INIT(SCALE_00, 0x1, 0x9, RATIO_DEF),
> + [7] = CH_DATA_INIT(SCALE_10, 0, 0, RATIO_DEF),
> + [9] = CH_DATA_INIT(SCALE_10, 0, 0, RATIO_DEF),
> + [10] = CH_DATA_INIT(SCALE_11, 0, 0, RATIO_DEF),
> + [11] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO_DEF),
> + [13] = CH_DATA_INIT(SCALE_01, 0, 0, RATIO_DEF),
> + [14] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO(68, 900)),
> + [15] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO(1, 3)),
> + [19] = CH_DATA_INIT(SCALE_11, 0, 0, RATIO_DEF),
> + [21] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO(3, 8)),
> + [22] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO(3, 8)),
> + [23] = CH_DATA_INIT(SCALE_00, 0, 0, RATIO(3, 8)),
> + [30] = CH_DATA_INIT(SCALE_11, 0, 0, RATIO_DEF),
> + [31] = CH_DATA_INIT(SCALE_11, 0, 0, RATIO_DEF),
> + };
> +
> + sprd_adc_ch_data_merge(data, ch_data, &ch_data_def);
> +}
> +
> +
> +static int sprd_adc_isen_enable(struct sprd_adc_data *data, int channel)
> +{
> + bool isen_support = data->ch_data[channel].isen_info & 0x1;
As below.
> +
> + if (!isen_support)
> + return 0;
> +
> + sprd_adc_regs_set_order(data, channel, REG_ISEN_ST, REG_ISEN_END);
> + usleep_range(500, 1000);
> +
> + return 0;
> +}
> +
> +static int sprd_adc_isen_disable(struct sprd_adc_data *data, int channel)
> +{
> + bool isen_support = data->ch_data[channel].isen_info & 0x1;
FIELD_GET()
> +
> + if (!isen_support)
as isen_support not otherwise used.
if (!FIELD_GET(...., ....))
> + return 0;
> +
> + sprd_adc_regs_clr_order_reverse(data, channel, REG_ISEN_ST, REG_ISEN_END);
> + usleep_range(500, 1000);
> +
> + return 0;
> +}
> +
> +
> +static int sprd_adc_enable(struct sprd_adc_data *data, int channel)
> +{
> + int ret = 0;
> + u32 reg_read = 0;
> +
> + if (data->pm_data.clk_regmap) {
> + ret = regmap_update_bits(data->pm_data.clk_regmap, data->pm_data.clk_reg,
> + data->pm_data.clk_reg_mask,
> + data->pm_data.clk_reg_mask);
> + ret |= regmap_read(data->pm_data.clk_regmap, data->pm_data.clk_reg, ®_read);
> + if (ret) {
> + dev_err(data->dev, "failed to enable clk26m, channel %d\n", channel);
> + return ret;
> + }
> + dev_dbg(data->dev, "enable clk26m: ch %d, reg_read 0x%x\n", channel, reg_read);
Directly accessing the regmap of a clock seems unusual. Why not provide generic clock interfaces
for this?
> + }
> +
> + ret = sprd_adc_isen_enable(data, channel);
> + if (ret) {
> + dev_err(data->dev, "failed to enable isen\n");
> + return ret;
> + }
> +
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_CTL,
> + SPRD_ADC_EN, SPRD_ADC_EN);
> + if (ret) {
> + dev_err(data->dev, "failed to set SPRD_ADC_EN\n");
> + return ret;
> + }
> +
> + return ret;
> +}
> +
> +static int sprd_adc_disable(struct sprd_adc_data *data, int channel)
> +{
> + int ret = 0;
Set just below here. Check for other instances of this.
> + u32 reg_read = 0;
> +
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_CTL, SPRD_ADC_EN, 0);
> + if (ret) {
> + dev_err(data->dev, "failed to reset SPRD_ADC_EN\n");
> + return ret;
> + }
> +
> + ret = sprd_adc_isen_disable(data, channel);
> + if (ret) {
> + dev_err(data->dev, "failed to disable isen\n");
> + return ret;
> + }
> +
> + if (data->pm_data.clk_regmap) {
> + ret = regmap_update_bits(data->pm_data.clk_regmap, data->pm_data.clk_reg,
> + data->pm_data.clk_reg_mask, 0);
> + ret |= regmap_read(data->pm_data.clk_regmap, data->pm_data.clk_reg, ®_read);
Don't combine ret values. Can result in complete garbage. Just handle each error
separately. It costs a few lines, but always works!
> + if (ret) {
> + dev_err(data->dev, "failed to disable clk26m, channel %d\n", channel);
> + return ret;
> + }
> + dev_dbg(data->dev, "disable clk26m: ch %d, reg_read 0x%x\n", channel, reg_read);
> + }
> +
> + return ret;
return 0;
Can't get here with any other value that I can see.
> +}
> +
> +static int sprd_adc_read(struct sprd_adc_data *data, int channel, int scale, int *val)
> +{
> + int ret = 0, sample_num_sw;
> + u32 rawdata = 0, tmp, status, scale_shift, scale_mask;
> + bool filter_sw = data->ch_data[channel].filter_info & 0x1;
> + int sample_num_hw = data->ch_data[channel].filter_info >> 1;
FIELD_GET() for both of these.
> +
> + if (data->pm_data.dev_suspended) {
> + dev_err(data->dev, "adc_exp: adc bas been suspended, ignore.\n");
has
> + return -EBUSY;
> + }
> +
> + dev_dbg(data->dev, "ch_data[%d]: scale %d, graph %d, filter_info 0x%x, isen_info 0x%x\n",
> + channel, data->ch_data[channel].scale, data->ch_data[channel].graph,
> + data->ch_data[channel].filter_info, data->ch_data[channel].isen_info);
> +
> + ret = hwspin_lock_timeout_raw(data->hwlock, SPRD_ADC_HWLOCK_TIMEOUT);
> + if (ret) {
> + dev_err(data->dev, "timeout to get the hwspinlock\n");
timeout getting the
or
timeout trying to get
> + return ret;
> + }
> +
> + ret = sprd_adc_enable(data, channel);
> + if (ret)
> + goto disable_adc;
> +
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_INT_CLR,
> + SPRD_ADC_IRQ_CLR, SPRD_ADC_IRQ_CLR);
> + if (ret)
> + goto disable_adc;
> +
> + /* Configure the channel id and scale */
> + scale_shift = data->var_data->reg_list[REG_SCALE].offset;
> + scale_mask = data->var_data->reg_list[REG_SCALE].mask;
> + tmp = (scale << scale_shift) & scale_mask;
> + tmp |= channel & SPRD_ADC_CHN_ID_MASK;
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_CH_CFG,
> + SPRD_ADC_CHN_ID_MASK | scale_mask, tmp);
> + if (ret)
> + goto disable_adc;
> +
> + /* Select 12bit conversion mode, and only sample 1 time */
> + tmp = SPRD_ADC_12BIT_MODE;
> + sample_num_sw = (filter_sw ? ADC_MESURE_NUMBER_SW - 1 : 0);
> + sample_num_hw = ((sample_num_hw > 0) ? sample_num_hw : ADC_MESURE_NUMBER_HW_DEF);
> + tmp |= ((unsigned long)sample_num_sw << SPRD_ADC_RUN_NUM_SHIFT) & SPRD_ADC_RUN_NUM_MASK;
> + tmp |= ((unsigned long)sample_num_hw << SPRD_ADC_AVERAGE_SHIFT) & SPRD_ADC_AVERAGE_MASK;
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_CTL,
> + SPRD_ADC_RUN_NUM_MASK | SPRD_ADC_12BIT_MODE |
> + SPRD_ADC_AVERAGE_MASK, tmp);
> + if (ret)
> + goto disable_adc;
> +
> + ret = regmap_update_bits(data->regmap, data->base + SPRD_ADC_CTL,
> + SPRD_ADC_CHN_RUN, SPRD_ADC_CHN_RUN);
> + if (ret)
> + goto disable_adc;
> +
> + ret = regmap_read_poll_timeout(data->regmap, data->base + SPRD_ADC_INT_RAW, status,
> + (status & SPRD_ADC_IRQ_RAW), SPRD_ADC_POLL_RAW_STATUS,
> + SPRD_ADC_RDY_TIMEOUT);
> + if (ret) {
> + dev_err(data->dev, "read adc timeout 0x%x\n", status);
> + sprd_adc_regs_dump(data, channel, scale, "t_bef");
> + sprd_adc_hw_enable(data);
> + sprd_adc_soft_rst(data);
> + sprd_adc_regs_dump(data, channel, scale, "t_aft");
> + goto disable_adc;
> + }
> +
> + if (filter_sw) {
> + rawdata = sprd_adc_get_val_with_sw_filter(data, channel);
> + } else {
> + ret = regmap_read(data->regmap, data->base + SPRD_ADC_DATA, &rawdata);
check ret.
> + rawdata &= SPRD_ADC_DATA_MASK;
> + }
> +disable_adc:
> + ret = sprd_adc_disable(data, channel);
> +
> + hwspin_unlock_raw(data->hwlock);
> +
> + if (!ret)
> + *val = rawdata;
I think val is valid even if the disable fails. So do it up a few lines unconditionally.
> +
> + return ret;
> +}
> +
> +static int sprd_adc_calculate_volt_by_graph(struct sprd_adc_data *data, int channel,
> + int scale, int raw_adc)
> +{
> + int tmp;
> + struct sprd_adc_graphs *graphs = &sprd_adc_graphs_array[data->var_data->graph_index];
> + int graph_offset = data->ch_data[channel].graph;
> + struct sprd_adc_linear_graph *linear_graph = &graphs->adc_graphs[graph_offset];
> +
> + tmp = (linear_graph->volt0 - linear_graph->volt1) * (raw_adc - linear_graph->adc1);
> + tmp /= (linear_graph->adc0 - linear_graph->adc1);
> + tmp += linear_graph->volt1;
> + tmp = (tmp < 0 ? 0 : tmp);
tmp = max(0, tmp) perhaps makes it clearer this is clamping a small negative (I think)
> +
> + dev_dbg(data->dev, "by_graph_c%d: v0 %d a0 %d, v1 %d a1 %d, raw_adc 0x%x, vol_graph %d\n",
> + channel, linear_graph->volt0, linear_graph->adc0, linear_graph->volt1,
> + linear_graph->adc1, raw_adc, tmp);
> +
> + return tmp;
> +}
> +
> +static int sprd_adc_calculate_volt_by_ratio(struct sprd_adc_data *data, int channel,
> + int scale, int vol_graph)
> +{
> + struct u32_fract ip_fract, aux_fract;
> + int pmic_type = data->var_data->pmic_type;
> + u32 vol_final = vol_graph;
> +
> + ratio_to_u32_fract(data->ch_data[channel].ip_ratio, &ip_fract);
> + vol_final = DIV_ROUND_CLOSEST(vol_final * ip_fract.denominator, ip_fract.numerator);
> +
> + ratio_to_u32_fract(data->ch_data[channel].ch_aux_ratio[scale], &aux_fract);
> + vol_final = DIV_ROUND_CLOSEST(vol_final * aux_fract.denominator, aux_fract.numerator);
> +
> + dev_dbg(data->dev, "by_ratio_c%d: id %d, scale %d, ip_r[%d/%d], aux_r[%d/%d], vol_f %d\n",
> + channel, pmic_type, scale, ip_fract.numerator, ip_fract.denominator,
> + aux_fract.numerator, aux_fract.denominator, vol_final);
> +
> + return vol_final;
> +}
> +
> +static int sprd_adc_read_processed(struct sprd_adc_data *data, int channel, int scale, int *val)
> +{
> + int ret, raw_adc, vol_graph;
> +
> + ret = sprd_adc_read(data, channel, scale, &raw_adc);
> +
> + if (ret)
> + return ret;
> +
> + vol_graph = sprd_adc_calculate_volt_by_graph(data, channel, scale, raw_adc);
> + *val = sprd_adc_calculate_volt_by_ratio(data, channel, scale, vol_graph);
> +
> + return 0;
> +}
> +
> +static int sprd_adc_ch_data_encode(struct sprd_adc_data *data, int ch)
> +{
> + int scale = data->ch_data[ch].scale & 0xff;
> + int isen_info = data->ch_data[ch].isen_info & 0xff;
> + int filter_info = data->ch_data[ch].filter_info & 0xff;
> +
> + return (scale | (filter_info << 8) | (isen_info << 16));
As below - looks like an array of u8. Treat it as such or use FIELD_PREP()
and treat it as a u32
> +}
> +
> +static void sprd_adc_ch_data_decode(struct sprd_adc_data *data, int ch, int val)
> +{
> + int scale_override, filter_override, isen_override;
> +
> + scale_override = (val & 0xff);
> + filter_override = ((val >> 8) & 0xff);
> + isen_override = ((val >> 16) & 0xff);
FIELD_GET() with appropriate mask defines.
Or just convert it to an array of u8 with appropriate endian conversion.
> +
> + if (scale_override < SPRD_ADC_SCALE_MAX && scale_override != data->ch_data[ch].scale)
> + data->ch_data[ch].scale = scale_override;
> +
> + if (data->ch_data[ch].filter_info != filter_override)
> + data->ch_data[ch].filter_info = filter_override;
> +
> + if (data->ch_data[ch].isen_info != isen_override)
> + data->ch_data[ch].isen_info = isen_override;
> +}
> +
...
> +static int sprd_adc_hw_enable(struct sprd_adc_data *data)
> +{
> + int ret;
> + u32 reg_addr, mask;
> +
> + reg_addr = GET_REG_ADDR(data, REG_XTL_CTRL);
I'm not sure the local variables much help readability.
So I'd just put them inline instead perhaps providing a
local variabel for the reglist.
u32 *reg_list = data->var_data->reg_list;
ret = regmap_update_bits(data->regmap,
GET_REG_ADDR(data, REG_XTL_CTRL),
reg_list[REG_XTL_CTRL]);
etc. May make sense for other similar cases.
> + mask = data->var_data->reg_list[REG_XTL_CTRL].mask;
> + ret = regmap_update_bits(data->regmap, reg_addr, mask, mask);
> + if (ret)
> + return ret;
> +
> + reg_addr = GET_REG_ADDR(data, REG_MODULE_EN);
> + mask = data->var_data->reg_list[REG_MODULE_EN].mask;
> + ret = regmap_update_bits(data->regmap, reg_addr, mask, mask);
> + if (ret)
> + return ret;
> +
> + /* Enable ADC work clock */
> + reg_addr = GET_REG_ADDR(data, REG_CLK_EN);
> + mask = data->var_data->reg_list[REG_CLK_EN].mask;
> + ret = regmap_update_bits(data->regmap, reg_addr, mask, mask);
> + if (ret)
> + return ret;
> +
> + return 0;
return regmap_update_bits()
> +}
> +
> +
> +static int sprd_adc_ch_data_init(struct sprd_adc_data *data)
> +{
> + struct device_node *np = data->dev->of_node;
> + int size, ret, ch, ch_data_val, i;
> + u32 *ch_data_override;
> +
> + data->var_data->ch_data_init(data);
> +
> + size = of_property_count_elems_of_size(np, "ch_data_override", sizeof(u32));
> + if (size <= 0)
> + return 0;
> +
> + if (size % 2) {
> + dev_err(data->dev, "Pair of ch data err!\n");
> + return -EINVAL;
> + }
> +
> + ch_data_override = kcalloc(size, sizeof(u32), GFP_KERNEL);
> + if (!ch_data_override)
> + return -ENOMEM;
> +
> + ret = of_property_read_u32_array(np, "ch_data_override", ch_data_override, size);
> + if (ret < 0) {
> + dev_err(data->dev, "Failed to read ch data from dt: %d\n", ret);
> + kfree(ch_data_override);
Probably better to use a goto for this cleanup so it can be shared with the normal
exit path. Of if you want to have fun (and don't need to backport this),
try the new __free() infrastructure to do scope based cleanup found in
linux/cleanup.h
This sort of pattern is exactly what that code is for.
> + return ret;
> + }
> +
> + for (i = 0; i < size; i += 2) {
> + ch = ch_data_override[i];
> + ch_data_val = ch_data_override[i+1];
kernel style always has spaces around an operator with two parameters.
[i + 1]
> + sprd_adc_ch_data_decode(data, ch, ch_data_val);
> + sprd_adc_ch_data_show(data, ch);
> + }
> +
> + kfree(ch_data_override);
> +
> + return 0;
> +}
> +
> +
> +static int sprd_adc_probe(struct platform_device *pdev)
> +{
> + struct device_node *np = pdev->dev.of_node;
> + struct sprd_adc_data *sprd_data;
> + const struct sprd_adc_variant_data *pdata;
> + struct iio_dev *indio_dev;
> + int ret;
> +
> + pdata = of_device_get_match_data(&pdev->dev);
device_get_match_data()
> + if (!pdata) {
> + dev_err(&pdev->dev, "No matching driver data found\n");
> + return -EINVAL;
> + }
> +
> + indio_dev = devm_iio_device_alloc(&pdev->dev, sizeof(*sprd_data));
> + if (!indio_dev)
> + return -ENOMEM;
> +
> + sprd_data = iio_priv(indio_dev);
> +
> + sprd_data->regmap = dev_get_regmap(pdev->dev.parent, NULL);
> + if (!sprd_data->regmap) {
> + dev_err(&pdev->dev, "failed to get ADC regmap\n");
> + return -ENODEV;
> + }
> +
> + ret = of_property_read_u32(np, "reg", &sprd_data->base);
Even though some elements of this (of_hwspin...) don't have generic firmware
interfaces, I would prefer to see those from linux/property.h used
wherever possible. It will take us a long time to make that a subsystem
wide change, but good not to have more unnecessary instances of device tree
specific property reading.
> + if (ret) {
> + dev_err(&pdev->dev, "failed to get ADC base address\n");
> + return ret;
> + }
> +
> + sprd_data->irq = platform_get_irq(pdev, 0);
> + if (sprd_data->irq < 0) {
> + dev_err(&pdev->dev, "failed to get ADC irq number\n");
> + return sprd_data->irq;
> + }
> +
> + ret = of_hwspin_lock_get_id(np, 0);
> + if (ret < 0) {
> + dev_err(&pdev->dev, "failed to get hwspinlock id\n");
> + return ret;
> + }
> +
> + mutex_init(&sprd_data->mutex);
> + sprd_data->hwlock = hwspin_lock_request_specific(ret);
> + if (!sprd_data->hwlock) {
> + dev_err(&pdev->dev, "failed to request hwspinlock\n");
> + return -ENXIO;
> + }
> +
> + ret = devm_add_action(&pdev->dev, sprd_adc_free_lock, sprd_data);
> + if (ret) {
> + sprd_adc_free_lock(sprd_data);
devm_add_action_or_reset() as below. Same for any more cases of this, I won't
mention it again (not I review up the driver as they make more logical sense
to me that way around ;)
> + dev_err(&pdev->dev, "failed to add free lock action\n");
> + return ret;
> + }
> +
> + sprd_data->dev = &pdev->dev;
> + sprd_data->var_data = pdata;
> + sprd_data->indio_dev = indio_dev;
> +
> + /* ADC channel scales calibration from nvmem device */
> + ret = sprd_adc_graphs_calibrate(sprd_data);
> + if (ret) {
> + dev_err(&pdev->dev, "failed to calib graphs from nvmem\n");
> + return ret;
> + }
> +
> + ret = sprd_adc_pm_init(sprd_data);
> + if (ret) {
> + dev_err(&pdev->dev, "adc pm init err.\n");
> + return ret;
> + }
> +
> + ret = sprd_adc_ch_data_init(sprd_data);
> + if (ret) {
> + dev_err(&pdev->dev, "ch data init err.\n");
> + return ret;
> + }
> +
> + ret = sprd_adc_hw_enable(sprd_data);
> + if (ret) {
> + dev_err(&pdev->dev, "failed to enable ADC module\n");
> + return ret;
> + }
> +
> + ret = devm_add_action(&pdev->dev, sprd_adc_hw_disable, sprd_data);
ret = devm_add_action_or_reset() and...
> + if (ret) {
> + sprd_adc_hw_disable(sprd_data);
drop this.
This is such a common case that we have the or_reset() form for it.
> + dev_err(&pdev->dev, "failed to add ADC disable action\n");
> + return ret;
return dev_err_probe(&pdev->dev, ret, "...");
Same for all similar error handling in probe, or functions that are only
called from probe.
> + }
> +
> + ret = sprd_adc_soft_rst(sprd_data);
> + if (ret) {
> + dev_err(&pdev->dev, "adc soft rst failed\n");
> + return ret;
return dev_err_probe(...);
> + }
> +
> + indio_dev->dev.parent = &pdev->dev;
> + indio_dev->name = dev_name(&pdev->dev);
> + indio_dev->modes = INDIO_DIRECT_MODE;
> + indio_dev->info = &sprd_info;
> + indio_dev->channels = sprd_channels;
> + indio_dev->num_channels = ARRAY_SIZE(sprd_channels);
> + ret = devm_iio_device_register(&pdev->dev, indio_dev);
> + if (ret)
> + dev_err(&pdev->dev, "could not register iio (ADC)");
Prefer return dev_err_probe() in all cases.
Also should not carry on to set drvdata if this failed as we should never
need it.
> +
> + platform_set_drvdata(pdev, indio_dev);
> +
> + return ret;
> +}
> +
> +static const struct sprd_adc_variant_data ump9620_data = {
> + .pmic_type = UMP9620_ADC,
> + .glb_reg_base = 0x2000,
> + .adc_reg_base_offset = 0x4,
> + .reg_list = regs_ump9620,
> + .graph_index = TWO_CELL_GRAPH,
> + .calib_infos = {
> + CALIB_INFO_INIT(11, SCALE_00, GRAPH_BIG),
> + CALIB_INFO_INIT(1, SCALE_00, GRAPH_SMALL),
> + CALIB_INFO_INIT(0, SCALE_01, GRAPH_VBAT_DET)
> + },
> + .aux_ratio_comm = {RATIO_DEF, RATIO(100, 133), RATIO(1000, 2600), RATIO(1000, 4060)},
> + .ch_data_init = ump9620_ch_data_init,
> +};
> +
> +static const struct sprd_adc_variant_data ump518_data = {
> + .pmic_type = UMP518_ADC,
> + .glb_reg_base = 0x1800,
> + .adc_reg_base_offset = 0x4,
> + .reg_list = regs_ump9620,
> + .graph_index = TWO_CELL_GRAPH,
> + .calib_infos = {
> + CALIB_INFO_INIT(11, SCALE_00, GRAPH_BIG),
> + CALIB_INFO_INIT(1, SCALE_00, GRAPH_SMALL),
> + CALIB_INFO_INIT(0, SCALE_01, GRAPH_VBAT_DET)
> + },
> + .aux_ratio_comm = {RATIO_DEF, RATIO(100, 133), RATIO(1000, 2600), RATIO(1000, 4060)},
> + .ch_data_init = ump9620_ch_data_init,
> +};
> +
> +static const struct of_device_id sprd_adc_of_match[] = {
> + { .compatible = "sprd,ump9620-adc", .data = &ump9620_data},
> + { .compatible = "sprd,ump518-adc", .data = &ump518_data},
Space before } for consistency with the opening bracket.
Numerical order preferred as for the bindings.
> + { }
> +};
> +
> +static const struct dev_pm_ops sprd_adc_pm_ops = {
> + .suspend_noirq = sprd_adc_pm_suspend,
> + .resume_noirq = sprd_adc_pm_resume,
> +};
> +
> +static struct platform_driver sprd_adc_driver = {
> + .probe = sprd_adc_probe,
> + .driver = {
> + .name = "sprd-adc",
> + .of_match_table = sprd_adc_of_match,
> + .pm = &sprd_adc_pm_ops,
Just space before = not a tab.
Also what stops this building without pm_sleep support?
You probably should either map to one of the macros that set the pm
ops up or add pm_sleep_ptr() protection to allow the sprd_adc_pm_ops
structure to be removed if appropriate.
> + },
> +};
> +
> +module_platform_driver(sprd_adc_driver);
> +MODULE_DESCRIPTION("SPRD PMIC ADC Driver");
> +MODULE_LICENSE("GPL");
