Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
IronPort-SDR: KoAwTB3paoivmfxqH9gQVk3qlUw62/68d7Bi9up/Z8+WykGTtll6JcXYadVWaO/+PAG+vWa67o
 8guaMaTC+SGaejW5t87Yv56qsvqTGNxeQ7m0p+N0OwQmo4jGNanMzBTJEHFhmYgh27JkHbMQzL
 Cx0oHdtT2Sbx7azFYmjc62gazieksZsL1uSjfEkxaMAHf0kAcdyEIkSlCHtiPxDiF9y6tAd7Kk
 nBOXp65HrEYH7/pPzveb77O1BcAgkYbp1xd3s6rJXLfiKNtPLL+IS3AX7w8xxzRLwNj43q+Mhb
 mO4=
References: <20200513125532.24585-1-lars.povlsen@microchip.com> <20200513125532.24585-13-lars.povlsen@microchip.com> <159054818459.88029.10644772284176356883@swboyd.mtv.corp.google.com>
From:   Lars Povlsen <lars.povlsen@microchip.com>
To:     Stephen Boyd <sboyd@kernel.org>
CC:     Arnd Bergmann <arnd@arndb.de>,
        Linus Walleij <linus.walleij@linaro.org>,
        SoC Team <soc@kernel.org>,
        Lars Povlsen <lars.povlsen@microchip.com>,
        "Steen Hegelund" <Steen.Hegelund@microchip.com>,
        Microchip Linux Driver Support <UNGLinuxDriver@microchip.com>,
        Olof Johansson <olof@lixom.net>,
        "Michael Turquette" <mturquette@baylibre.com>,
        <devicetree@vger.kernel.org>, <linux-clk@vger.kernel.org>,
        <linux-gpio@vger.kernel.org>,
        <linux-arm-kernel@lists.infradead.org>,
        <linux-kernel@vger.kernel.org>,
        Alexandre Belloni <alexandre.belloni@bootlin.com>
Subject: Re: [PATCH 12/14] clk: sparx5: Add Sparx5 SoC DPLL clock driver
In-Reply-To: <159054818459.88029.10644772284176356883@swboyd.mtv.corp.google.com>
Date:   Wed, 27 May 2020 16:29:17 +0200
Message-ID: <87mu5tpi2q.fsf@soft-dev15.microsemi.net>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


Stephen Boyd writes:

> Quoting Lars Povlsen (2020-05-13 05:55:30)
>> diff --git a/drivers/clk/clk-sparx5.c b/drivers/clk/clk-sparx5.c
>> new file mode 100644
>> index 0000000000000..685b3028a7071
>> --- /dev/null
>> +++ b/drivers/clk/clk-sparx5.c
>> @@ -0,0 +1,269 @@
>> +// SPDX-License-Identifier: GPL-2.0-or-later
>> +/*
>> + * Microchip Sparx5 SoC Clock driver.
>> + *
>> + * Copyright (c) 2019 Microchip Inc.
>> + *
>> + * Author: Lars Povlsen <lars.povlsen@microchip.com>
>> + */
>> +
>> +#include <linux/io.h>
>> +#include <linux/clk-provider.h>
>> +#include <linux/of.h>
>> +#include <linux/of_address.h>
>> +#include <linux/slab.h>
>> +#include <linux/platform_device.h>
>> +#include <dt-bindings/clock/microchip,sparx5.h>
>> +
>> +#define PLL_DIV_MASK           GENMASK(7, 0)
>> +#define PLL_PRE_DIV_MASK       GENMASK(10, 8)
>> +#define PLL_PRE_DIV_SHIFT      8
>> +#define PLL_ROT_DIR            BIT(11)
>> +#define PLL_ROT_SEL_MASK       GENMASK(13, 12)
>> +#define PLL_ROT_SEL_SHIFT      12
>> +#define PLL_ROT_ENA            BIT(14)
>> +#define PLL_CLK_ENA            BIT(15)
>> +
>> +#define MAX_SEL 4
>> +#define MAX_PRE BIT(3)
>> +
>> +#define KHZ 1000
>> +#define MHZ (KHZ*KHZ)
>
> I suspect (1000 * KHZ) would make more sense.
>

Fine.

>> +
>> +#define BASE_CLOCK (2500UL*MHZ)
>> +
>> +static u8 sel_rates[MAX_SEL] = { 0, 2*8, 2*4, 2*2 };
>
> const?
>

Yes, sure.

>> +
>> +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;
>> +};
>> +
>> +struct s5_clk_data {
>> +       void __iomem *base;
>> +       struct s5_hw_clk s5_hw[N_CLOCKS];
>> +};
>> +
>> +struct pll_conf {
>> +       int freq;
>> +       u8 div;
>> +       bool rot_ena;
>> +       u8 rot_sel;
>> +       u8 rot_dir;
>> +       u8 pre_div;
>> +};
>> +
>> +#define to_clk_pll(hw) container_of(hw, struct s5_hw_clk, hw)
>> +
>> +unsigned long calc_freq(const struct pll_conf *pdata)
>> +{
>> +       unsigned long rate = BASE_CLOCK / pdata->div;
>> +
>> +       if (pdata->rot_ena) {
>> +               unsigned long base = BASE_CLOCK / pdata->div;
>> +               int sign = pdata->rot_dir ? -1 : 1;
>> +               int divt = sel_rates[pdata->rot_sel] * (1 + pdata->pre_div);
>> +               int divb = divt + sign;
>> +
>> +               rate = mult_frac(base, divt, divb);
>> +               rate = roundup(rate, 1000);
>> +       }
>> +
>> +       return rate;
>> +}
>> +
>> +static unsigned long clk_calc_params(unsigned long rate,
>> +                                    struct pll_conf *conf)
>> +{
>> +       memset(conf, 0, sizeof(*conf));
>> +
>> +       conf->div = DIV_ROUND_CLOSEST_ULL(BASE_CLOCK, rate);
>> +
>> +       if (BASE_CLOCK % rate) {
>> +               struct pll_conf best;
>> +               ulong cur_offset, best_offset = rate;
>> +               int i, j;
>> +
>> +               /* Enable fractional rotation */
>> +               conf->rot_ena = true;
>> +
>> +               if ((BASE_CLOCK / rate) != conf->div) {
>> +                       /* Overshoot, adjust other direction */
>> +                       conf->rot_dir = 1;
>> +               }
>> +
>> +               /* Brute force search over MAX_PRE * (MAX_SEL - 1) = 24 */
>> +               for (i = 0; i < MAX_PRE; i++) {
>> +                       conf->pre_div = i;
>> +                       for (j = 1; j < MAX_SEL; j++) {
>> +                               conf->rot_sel = j;
>> +                               conf->freq = calc_freq(conf);
>> +                               cur_offset = abs(rate - conf->freq);
>> +                               if (cur_offset == 0)
>> +                                       /* Perfect fit */
>> +                                       goto done;
>
> Why not 'break' and drop the label?
>

Its a dual loop. Anyway, I changed it to add "best_offset > 0" in the
loop guards and drop "cur_offset == 0" as a special case, so no goto.

>> +                               if (cur_offset < best_offset) {
>> +                                       /* Better fit found */
>> +                                       best_offset = cur_offset;
>> +                                       best = *conf;
>> +                               }
>> +                       }
>> +               }
>> +               /* Best match */
>> +               *conf = best;
>> +       }
>> +
>> +done:
>> +       return conf->freq;
>> +}
>> +
>> +static int clk_pll_enable(struct clk_hw *hw)
>> +{
>> +       struct s5_hw_clk *pll = to_clk_pll(hw);
>> +       u32 val = readl(pll->reg);
>> +
>> +       val |= PLL_CLK_ENA;
>> +       writel(val, pll->reg);
>> +       pr_debug("%s: Enable val %04x\n", clk_names[pll->index], val);
>> +       return 0;
>> +}
>> +
>> +static void clk_pll_disable(struct clk_hw *hw)
>> +{
>> +       struct s5_hw_clk *pll = to_clk_pll(hw);
>> +       u32 val = readl(pll->reg);
>> +
>> +       val &= ~PLL_CLK_ENA;
>> +       writel(val, pll->reg);
>> +       pr_debug("%s: Disable val %04x\n", clk_names[pll->index], val);
>
> Can we drop these pr_debug() prints? They're probably never going to be
> used after developing this driver.
>
>> +}
>> +
>> +static int clk_pll_set_rate(struct clk_hw *hw,
>
> Please rename clk_pll to something less generic, like s5_pll or
> something.
>

Yeah, I see that. I changed all generic symbols to use s5_ prefix where
applicable. Also fixed non-static calc_freq() symbol.

>> +                           unsigned long rate,
>> +                           unsigned long parent_rate)
>> +{
>> +       struct s5_hw_clk *pll = to_clk_pll(hw);
>> +       struct pll_conf conf;
>> +       unsigned long eff_rate;
>> +       int ret = 0;
>> +
>> +       eff_rate = clk_calc_params(rate, &conf);
>> +       if (eff_rate == rate) {
>> +               u32 val;
>> +
>> +               val = readl(pll->reg) & PLL_CLK_ENA;
>> +               val |= PLL_DIV_MASK & conf.div;
>> +               if (conf.rot_ena) {
>> +                       val |= (PLL_ROT_ENA |
>> +                               (PLL_ROT_SEL_MASK &
>> +                                (conf.rot_sel << PLL_ROT_SEL_SHIFT)) |
>> +                               (PLL_PRE_DIV_MASK &
>> +                                (conf.pre_div << PLL_PRE_DIV_SHIFT)));
>
> This can use the FIELD_GET and helpers?
>

Yes, makes sense. Done.

>> +                       if (conf.rot_dir)
>> +                               val |= PLL_ROT_DIR;
>> +               }
>> +               pr_debug("%s: Rate %ld >= 0x%04x\n",
>> +                        clk_names[pll->index], rate, val);
>> +               writel(val, pll->reg);
>> +       } else {
>> +               pr_err("%s: freq unsupported: %ld paren %ld\n",
>> +                      clk_names[pll->index], rate, parent_rate);
>> +               ret = -ENOTSUPP;
>
> I'd prefer we short circuit the function
>
>         eff_rate = clk_calc_params(...);
>         if (eff_rate != rate)
>                 return -ENOTSUPP;
>
>         do the other things...
>
> This avoids lots of indentation.

Ok, noted.

>
>> +       }
>> +
>> +       return ret;
>> +}
>> +
>> +static unsigned long clk_pll_recalc_rate(struct clk_hw *hw,
>> +                                        unsigned long parent_rate)
>> +{
>> +       /* Don't care */
>
> What does this mean? recalc_rate is supposed to tell us what rate has
> been achieved for this clk.

I added a proper implementation for this.

>
>> +       return 0;
>> +}
>> +
>> +static long clk_pll_round_rate(struct clk_hw *hw, unsigned long rate,
>> +                              unsigned long *parent_rate)
>> +{
>> +       struct pll_conf conf;
>> +       unsigned long eff_rate;
>> +
>> +       eff_rate = clk_calc_params(rate, &conf);
>> +       pr_debug("%s: Rate %ld rounded to %ld\n", __func__, rate, eff_rate);
>> +
>> +       return eff_rate;
>> +}
>> +
>> +static const struct clk_ops s5_pll_ops = {
>> +       .enable         = clk_pll_enable,
>> +       .disable        = clk_pll_disable,
>> +       .set_rate       = clk_pll_set_rate,
>> +       .round_rate     = clk_pll_round_rate,
>> +       .recalc_rate    = clk_pll_recalc_rate,
>> +};
>> +
>> +static struct s5_clk_data *s5_clk_alloc(struct device_node *np)
>> +{
>> +       struct s5_clk_data *clk_data;
>> +
>> +       clk_data = kzalloc(sizeof(*clk_data), GFP_KERNEL);
>> +       if (WARN_ON(!clk_data))
>
> Drop the WARN_ON(), kzalloc() already prints a big stacktrace when it
> fails.

Yes.

>
>> +               return NULL;
>> +
>> +       clk_data->base = of_iomap(np, 0);
>> +       if (WARN_ON(!clk_data->base))
>> +               return NULL;
>> +
>> +       return clk_data;
>
> Just inline this function at the callsite please.
>

Yes.

>> +}
>> +
>> +static struct clk_hw *s5_clk_hw_get(struct of_phandle_args *clkspec, void *data)
>> +{
>> +       struct s5_clk_data *pll_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 &pll_clk->s5_hw[idx].hw;
>> +}
>> +
>> +static void __init s5_pll_init(struct device_node *np)
>> +{
>> +       int i, ret;
>> +       struct s5_clk_data *pll_clk;
>> +       struct clk_init_data init = { 0 };
>
> Just do init = { } so that 0 doesn't trip up sparse.

I'm not sure what you mean by "trip up sparse", but its changed now.

>
>> +
>> +       pll_clk = s5_clk_alloc(np);
>> +       if (!pll_clk)
>> +               return;
>> +
>> +       init.ops = &s5_pll_ops;
>> +       init.parent_names = NULL;
>> +       init.num_parents = 0;
>
> Drop these last two lines if there aren't any parents.
>

OK.

>> +
>> +       for (i = 0; i < N_CLOCKS; i++) {
>> +               struct s5_hw_clk *s5_hw = &pll_clk->s5_hw[i];
>> +
>> +               init.name = clk_names[i];
>> +               s5_hw->index = i;
>> +               s5_hw->reg = pll_clk->base + (i * sizeof(u32));
>> +               s5_hw->hw.init = &init;
>> +               ret = of_clk_hw_register(np, &s5_hw->hw);
>> +               if (ret) {
>> +                       pr_err("failed to register %s clock\n", init.name);
>> +                       return;
>> +               }
>> +       }
>> +
>> +       of_clk_add_hw_provider(np, s5_clk_hw_get, pll_clk);
>> +}
>> +CLK_OF_DECLARE_DRIVER(microchip_s5, "microchip,sparx5-dpll", s5_pll_init);
>
> Why DECLARE_DRIVER? Please add a comment indicating the other driver
> that is supposed to probe against this node. And is there any reason
> this can't be a platform driver? That is preferred over
> CLK_OF_DECLARE*() usage.

I will change it to a platform driver.

Thank you very much for your comments, they are highly appreciated.

---Lars

--
Lars Povlsen,
Microchip