Hi Stephen!
Thank you for your comments!
Stephen Boyd writes:
> Quoting Lars Povlsen (2020-06-15 06:32:40)
>> diff --git a/drivers/clk/clk-sparx5.c b/drivers/clk/clk-sparx5.c
>> new file mode 100644
>> index 0000000000000..c2e7aa0214ebd
>> --- /dev/null
>> +++ b/drivers/clk/clk-sparx5.c
>> @@ -0,0 +1,312 @@
>> +// SPDX-License-Identifier: GPL-2.0-or-later
>> +/*
>> + * Microchip Sparx5 SoC Clock driver.
>> + *
>> + * Copyright (c) 2019 Microchip Inc.
>> + *
>> + * Author: Lars Povlsen <lars.povlsen@xxxxxxxxxxxxx>
>> + */
>> +
>> +#include <linux/io.h>
>> +#include <linux/module.h>
>> +#include <linux/clk-provider.h>
>> +#include <linux/bitfield.h>
>> +#include <linux/of.h>
>> +#include <linux/of_address.h>
>
> Is this include used?
Nope, gone now.
>
>> +#include <linux/slab.h>
>> +#include <linux/platform_device.h>
>> +#include <dt-bindings/clock/microchip,sparx5.h>
>> +
>> +#define PLL_DIV GENMASK(7, 0)
>> +#define PLL_PRE_DIV GENMASK(10, 8)
>> +#define PLL_ROT_DIR BIT(11)
>> +#define PLL_ROT_SEL GENMASK(13, 12)
>> +#define PLL_ROT_ENA BIT(14)
>> +#define PLL_CLK_ENA BIT(15)
>> +
>> +#define MAX_SEL 4
>> +#define MAX_PRE BIT(3)
>> +
>> +static const u8 sel_rates[MAX_SEL] = { 0, 2*8, 2*4, 2*2 };
>> +
>> +static const char *clk_names[N_CLOCKS] = {
>> + "core", "ddr", "cpu2", "arm2",
>> + "aux1", "aux2", "aux3", "aux4",
>> + "synce",
>> +};
>> +
>> +struct s5_hw_clk {
>> + struct clk_hw hw;
>> + void __iomem *reg;
>> + int index;
>
> This looks unused. Drop it?
>
Yes, that was a leftover. Good find.
>> +};
>> +
>> +struct s5_clk_data {
>> + void __iomem *base;
>> + struct s5_hw_clk s5_hw[N_CLOCKS];
>> +};
>> +
>> +struct s5_pll_conf {
>> + int freq;
>
> Why not unsigned long like the type that s5_calc_freq() returns?
>
Good point, changed it.
>> + u8 div;
>> + bool rot_ena;
>> + u8 rot_sel;
>> + u8 rot_dir;
>> + u8 pre_div;
>> +};
>> +
>> +#define to_s5_pll(hw) container_of(hw, struct s5_hw_clk, hw)
>> +
>> +static unsigned long s5_calc_freq(unsigned long parent_rate,
>> + const struct s5_pll_conf *conf)
>> +{
>> + unsigned long rate = parent_rate / conf->div;
>> +
>> + if (conf->rot_ena) {
>> + int sign = conf->rot_dir ? -1 : 1;
>> + int divt = sel_rates[conf->rot_sel] * (1 + conf->pre_div);
>> + int divb = divt + sign;
>> +
>> + rate = mult_frac(rate, divt, divb);
>> + rate = roundup(rate, 1000);
>> + }
>> +
>> + return rate;
>> +}
>> +
>> +static void s5_search_fractional(unsigned long rate,
>> + unsigned long parent_rate,
>> + int div,
>> + struct s5_pll_conf *conf)
>> +{
>> + struct s5_pll_conf best;
>> + ulong cur_offset, best_offset = rate;
>> + int d, i, j;
>> +
>> + memset(conf, 0, sizeof(*conf));
>> + conf->div = div;
>> + conf->rot_ena = 1; /* Fractional rate */
>> +
>> + for (d = 0; best_offset > 0 && d <= 1 ; d++) {
>> + conf->rot_dir = !!d;
>> + for (i = 0; best_offset > 0 && i < MAX_PRE; i++) {
>> + conf->pre_div = i;
>> + for (j = 1; best_offset > 0 && j < MAX_SEL; j++) {
>> + conf->rot_sel = j;
>> + conf->freq = s5_calc_freq(parent_rate, conf);
>> + cur_offset = abs(rate - conf->freq);
>> + if (cur_offset < best_offset) {
>> + best_offset = cur_offset;
>> + best = *conf;
>> + }
>> + }
>> + }
>> + }
>> +
>> + /* Best match */
>> + *conf = best;
>> +}
>> +
>> +static unsigned long s5_calc_params(unsigned long rate,
>> + unsigned long parent_rate,
>> + struct s5_pll_conf *conf)
>> +{
>> + if (parent_rate % rate) {
>> + struct s5_pll_conf alt1, alt2;
>> + int div;
>> +
>> + div = DIV_ROUND_CLOSEST_ULL(parent_rate, rate);
>> + s5_search_fractional(rate, parent_rate, div, &alt1);
>> +
>> + /* Straight match? */
>> + if (alt1.freq == rate) {
>> + *conf = alt1;
>> + } else {
>> + /* Try without rounding divider */
>> + div = parent_rate / rate;
>> + if (div != alt1.div) {
>> + s5_search_fractional(rate, parent_rate, div,
>> + &alt2);
>> + /* Select the better match */
>> + if (abs(rate - alt1.freq) <
>> + abs(rate - alt2.freq))
>> + *conf = alt1;
>> + else
>> + *conf = alt2;
>> + }
>> + }
>> + } else {
>> + /* Straight fit */
>> + memset(conf, 0, sizeof(*conf));
>> + conf->div = parent_rate / rate;
>> + }
>> +
>> + return conf->freq;
>> +}
>> +
>> +static int s5_pll_enable(struct clk_hw *hw)
>> +{
>> + struct s5_hw_clk *pll = to_s5_pll(hw);
>> + u32 val = readl(pll->reg);
>> +
>> + val |= PLL_CLK_ENA;
>> + writel(val, pll->reg);
>> +
>> + return 0;
>> +}
>> +
>> +static void s5_pll_disable(struct clk_hw *hw)
>> +{
>> + struct s5_hw_clk *pll = to_s5_pll(hw);
>> + u32 val = readl(pll->reg);
>> +
>> + val &= ~PLL_CLK_ENA;
>> + writel(val, pll->reg);
>> +}
>> +
>> +static int s5_pll_set_rate(struct clk_hw *hw,
>> + unsigned long rate,
>> + unsigned long parent_rate)
>> +{
>> + struct s5_hw_clk *pll = to_s5_pll(hw);
>> + struct s5_pll_conf conf;
>> + unsigned long eff_rate;
>> + u32 val;
>> +
>> + eff_rate = s5_calc_params(rate, parent_rate, &conf);
>> + if (eff_rate != rate)
>> + return -EOPNOTSUPP;
>> +
>> + val = readl(pll->reg) & PLL_CLK_ENA;
>> + val |= FIELD_PREP(PLL_DIV, conf.div);
>> + if (conf.rot_ena) {
>> + val |= PLL_ROT_ENA;
>> + val |= FIELD_PREP(PLL_ROT_SEL, conf.rot_sel);
>> + val |= FIELD_PREP(PLL_PRE_DIV, conf.pre_div);
>> + if (conf.rot_dir)
>> + val |= PLL_ROT_DIR;
>> + }
>> + writel(val, pll->reg);
>> +
>> + return 0;
>> +}
>> +
>> +static unsigned long s5_pll_recalc_rate(struct clk_hw *hw,
>> + unsigned long parent_rate)
>> +{
>> + struct s5_hw_clk *pll = to_s5_pll(hw);
>> + struct s5_pll_conf conf;
>> + u32 val;
>> +
>> + val = readl(pll->reg);
>> +
>> + if (val & PLL_CLK_ENA) {
>> + conf.div = FIELD_GET(PLL_DIV, val);
>> + conf.pre_div = FIELD_GET(PLL_PRE_DIV, val);
>> + conf.rot_ena = FIELD_GET(PLL_ROT_ENA, val);
>> + conf.rot_dir = FIELD_GET(PLL_ROT_DIR, val);
>> + conf.rot_sel = FIELD_GET(PLL_ROT_SEL, val);
>> +
>> + conf.freq = s5_calc_freq(parent_rate, &conf);
>> + } else
>> + conf.freq = 0;
>
> Nitpick: Please add braces on single line else statements when the if is
> multiline.
>
Done.
>> +
>> + return conf.freq;
>> +}
>> +
>> +static long s5_pll_round_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long *parent_rate)
>> +{
>> + struct s5_pll_conf conf;
>> +
>> + return s5_calc_params(rate, *parent_rate, &conf);
>> +}
>> +
>> +static const struct clk_ops s5_pll_ops = {
>> + .enable = s5_pll_enable,
>> + .disable = s5_pll_disable,
>> + .set_rate = s5_pll_set_rate,
>> + .round_rate = s5_pll_round_rate,
>> + .recalc_rate = s5_pll_recalc_rate,
>> +};
>> +
>> +static struct clk_hw *s5_clk_hw_get(struct of_phandle_args *clkspec, void *data)
>> +{
>> + struct s5_clk_data *s5_clk = data;
>> + unsigned int idx = clkspec->args[0];
>> +
>> + if (idx >= N_CLOCKS) {
>> + pr_err("%s: invalid index %u\n", __func__, idx);
>> + return ERR_PTR(-EINVAL);
>> + }
>> +
>> + return &s5_clk->s5_hw[idx].hw;
>> +}
>> +
>> +static int s5_clk_probe(struct platform_device *pdev)
>> +{
>> + struct device *dev = &pdev->dev;
>> + struct device_node *np = dev->of_node;
>> + int i, ret;
>> + struct s5_clk_data *s5_clk;
>> + const char *parent_name;
>> + struct clk_init_data init = {
>> + .ops = &s5_pll_ops,
>> + .parent_names = &parent_name,
>
> It looks like with the binding you can drop parent_name and just use
> .parent_data = { .index = 0 } and event drop the .index = 0 bit because
> that's the default valjue.
>
Ok, I got rid of the parent_name and used .parent_data instead.
>> + .num_parents = 1,
>> + };
>> +
>> + s5_clk = devm_kzalloc(dev, sizeof(*s5_clk), GFP_KERNEL);
>> + if (!s5_clk)
>> + return -ENOMEM;
>> +
>> + s5_clk->base = devm_platform_ioremap_resource(pdev, 0);
>> + if (IS_ERR(s5_clk->base))
>> + return PTR_ERR(s5_clk->base);
>> +
>> + parent_name = of_clk_get_parent_name(np, 0);
>> + if (!parent_name) {
>> + dev_err(dev, "%pOFn: missing parent clock\n", np);
>> + return -EINVAL;
>> + }
>
> It's nice because then this call goes away.
>
Yes.
>> +
>> + for (i = 0; i < N_CLOCKS; i++) {
>> + struct s5_hw_clk *s5_hw = &s5_clk->s5_hw[i];
>> +
>> + init.name = clk_names[i];
>> + s5_hw->index = i;
>> + s5_hw->reg = s5_clk->base + (i * sizeof(u32));
>
> I'd prefer i * 4 because the hardware engineers probably don't care
> about the size of a u32 in bytes.
>
Sure.
>> + s5_hw->hw.init = &init;
>> + ret = devm_clk_hw_register(dev, &s5_hw->hw);
>> + if (ret) {
>> + dev_err(dev, "failed to register %s clock\n",
>> + init.name);
>
> init.name will be destroyed. Just drop this error message? Maybe we
> should add it into the core framework because quite a few driver authors
> want it.
>
>> + return ret;
>> + }
>> + }
>> +
>> + return of_clk_add_hw_provider(np, s5_clk_hw_get, s5_clk);
>
> Use devm_of_clk_add_hw_provider()?
Yes. Also removes 'np' variable.
>
>> +}
>> +
>> +static int s5_clk_remove(struct platform_device *pdev)
>> +{
>> + of_clk_del_provider(pdev->dev.of_node);
>> +
>> + return 0;
>> +}
>> +
>
> And then remove this whole remove function?
>
And yes.
>> +static const struct of_device_id s5_clk_dt_ids[] = {
>> + { .compatible = "microchip,sparx5-dpll", },
>> + { }
>> +};
>> +MODULE_DEVICE_TABLE(of, s5_clk_dt_ids);
>> +
>> +static struct platform_driver s5_clk_driver = {
>> + .probe = s5_clk_probe,
>> + .remove = s5_clk_remove,
>> + .driver = {
>> + .name = "sparx5-clk",
>> + .of_match_table = s5_clk_dt_ids,
>> + },
>> +};
Thanks a lot for your comments - that shaved off some code!
I'll be refreshing the series shortly.
Cheers,
--
Lars Povlsen,
Microchip
