On Tue, Mar 21, 2023 at 4:06 AM Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
>
> Quoting Keguang Zhang (2023-03-16 03:47:06)
> > diff --git a/drivers/clk/clk-loongson1.c b/drivers/clk/clk-loongson1.c
> > new file mode 100644
> > index 000000000000..4fda55c67d8d
> > --- /dev/null
> > +++ b/drivers/clk/clk-loongson1.c
> > @@ -0,0 +1,301 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Clock driver for Loongson-1 SoC
> > + *
> > + * Copyright (C) 2012-2023 Keguang Zhang <keguang.zhang@xxxxxxxxx>
> > + */
> > +
> > +#include <linux/clk-provider.h>
> > +#include <linux/of_address.h>
> > +#include <linux/slab.h>
>
> Need some more includes here, for container_of() and GENMASK(), readl(),
> etc.
>
Will do.
> > +
> > +#include <dt-bindings/clock/loongson,ls1x-clk.h>
> > +
> > +/* Loongson 1 Clock Register Definitions */
> > +#define CLK_PLL_FREQ 0x0
> > +#define CLK_PLL_DIV 0x4
> > +
> > +static DEFINE_SPINLOCK(ls1x_clk_div_lock);
> > +
>
> Needs include.
>
Will do.
> > +struct ls1x_clk_pll_data {
> > + u32 fixed;
> > + u8 shift;
> > + u8 int_shift;
> > + u8 int_width;
> > + u8 frac_shift;
> > + u8 frac_width;
> > +};
> > +
> > +struct ls1x_clk_div_data {
> > + u8 shift;
> > + u8 width;
> > + unsigned long flags;
> > + const struct clk_div_table *table;
> > + u8 bypass_shift;
> > + u8 bypass_inv;
> > + spinlock_t *lock; /* protect access to DIV registers */
> > +};
> > +
> > +struct ls1x_clk {
> > + void __iomem *reg;
> > + unsigned int offset;
> > + struct clk_hw hw;
> > + void *data;
> > +};
> > +
> > +#define to_ls1x_clk(_hw) container_of(_hw, struct ls1x_clk, hw)
> > +
> > +static inline unsigned long ls1x_pll_rate_part(unsigned int val,
>
> return a u32?
>
Yes.
> > + unsigned int shift,
> > + unsigned int width)
> > +{
> > + return (val & GENMASK(shift + width, shift)) >> shift;
> > +}
> > +
> > +static unsigned long ls1x_pll_recalc_rate(struct clk_hw *hw,
> > + unsigned long parent_rate)
> > +{
> > + struct ls1x_clk *ls1x_clk = to_ls1x_clk(hw);
> > + const struct ls1x_clk_pll_data *d = ls1x_clk->data;
> > + u32 val, rate;
> > +
> > + val = readl(ls1x_clk->reg);
> > + rate = d->fixed;
> > + rate += ls1x_pll_rate_part(val, d->int_shift, d->int_width);
> > + if (d->frac_width)
> > + rate += ls1x_pll_rate_part(val, d->frac_shift, d->frac_width);
> > + rate *= parent_rate;
> > + rate >>= d->shift;
> > +
> > + return rate;
> > +}
> > +
> > +static const struct clk_ops ls1x_pll_clk_ops = {
> > + .recalc_rate = ls1x_pll_recalc_rate,
> > +};
> > +
> > +static unsigned long ls1x_divider_recalc_rate(struct clk_hw *hw,
> > + unsigned long parent_rate)
> > +{
> > + struct ls1x_clk *ls1x_clk = to_ls1x_clk(hw);
> > + const struct ls1x_clk_div_data *d = ls1x_clk->data;
> > + unsigned int val;
> > +
> > + val = readl(ls1x_clk->reg) >> d->shift;
> > + val &= clk_div_mask(d->width);
> > +
> > + return divider_recalc_rate(hw, parent_rate, val, d->table,
> > + d->flags, d->width);
> > +}
> > +
> > +static long ls1x_divider_round_rate(struct clk_hw *hw, unsigned long rate,
> > + unsigned long *prate)
> > +{
> > + struct ls1x_clk *ls1x_clk = to_ls1x_clk(hw);
> > + const struct ls1x_clk_div_data *d = ls1x_clk->data;
> > +
> > + return divider_round_rate(hw, rate, prate, d->table,
> > + d->width, d->flags);
> > +}
> > +
> > +static int ls1x_divider_set_rate(struct clk_hw *hw, unsigned long rate,
> > + unsigned long parent_rate)
> > +{
> > + struct ls1x_clk *ls1x_clk = to_ls1x_clk(hw);
> > + const struct ls1x_clk_div_data *d = ls1x_clk->data;
> > + int val, div_val;
> > + unsigned long flags = 0;
> > +
> > + div_val = divider_get_val(rate, parent_rate, d->table,
> > + d->width, d->flags);
> > + if (div_val < 0)
> > + return div_val;
> > +
> > + if (d->lock)
> > + spin_lock_irqsave(d->lock, flags);
> > + else
> > + __acquire(d->lock);
> > +
> > + /* Bypass the clock */
> > + val = readl(ls1x_clk->reg);
> > + if (d->bypass_inv)
> > + val &= ~BIT(d->bypass_shift);
> > + else
> > + val |= BIT(d->bypass_shift);
> > + writel(val, ls1x_clk->reg);
> > +
> > + val = readl(ls1x_clk->reg);
> > + val &= ~(clk_div_mask(d->width) << d->shift);
> > + val |= (u32)div_val << d->shift;
> > + writel(val, ls1x_clk->reg);
> > +
> > + /* Restore the clock */
> > + val = readl(ls1x_clk->reg);
> > + if (d->bypass_inv)
> > + val |= BIT(d->bypass_shift);
> > + else
> > + val &= ~BIT(d->bypass_shift);
> > + writel(val, ls1x_clk->reg);
> > +
> > + if (d->lock)
> > + spin_unlock_irqrestore(d->lock, flags);
> > + else
> > + __release(d->lock);
>
> Is there a case where there isn't a lock? It would be easier to read if
> this always had a lock and it wasn't optional.
>
Indeed. The lock always exists in this driver.
Will remove the unnecessary __acquire()/__release().
> > +
> > + return 0;
> > +}
> > +
> > +static const struct clk_ops ls1x_clk_divider_ops = {
> > + .recalc_rate = ls1x_divider_recalc_rate,
> > + .round_rate = ls1x_divider_round_rate,
> > + .set_rate = ls1x_divider_set_rate,
> > +};
> > +
> > +#define LS1X_CLK_PLL(_name, _offset, _fixed, _shift, \
> > + f_shift, f_width, i_shift, i_width) \
> > +struct ls1x_clk _name = { \
> > + .offset = (_offset), \
> > + .data = &(struct ls1x_clk_pll_data) { \
> > + .fixed = (_fixed), \
> > + .shift = (_shift), \
> > + .int_shift = (i_shift), \
> > + .int_width = (i_width), \
> > + .frac_shift = (f_shift), \
> > + .frac_width = (f_width), \
> > + }, \
> > + .hw.init = &(struct clk_init_data) { \
> > + .name = #_name, \
> > + .ops = &ls1x_pll_clk_ops, \
> > + .parent_data = &(const struct clk_parent_data) { \
> > + .fw_name = "xtal", \
> > + .name = "xtal", \
> > + .index = -1, \
> > + }, \
> > + .num_parents = 1, \
> > + }, \
> > +}
> > +
> > +#define LS1X_CLK_DIV(_name, _pname, _offset, _shift, _width, \
> > + _table, _bypass_shift, _bypass_inv, _flags) \
> > +struct ls1x_clk _name = { \
> > + .offset = (_offset), \
> > + .data = &(struct ls1x_clk_div_data){ \
> > + .shift = (_shift), \
> > + .width = (_width), \
> > + .table = (_table), \
> > + .flags = (_flags), \
> > + .bypass_shift = (_bypass_shift), \
> > + .bypass_inv = (_bypass_inv), \
> > + .lock = &ls1x_clk_div_lock, \
> > + }, \
> > + .hw.init = &(struct clk_init_data) { \
>
> Can be const.
>
Will do.
> > + .name = #_name, \
> > + .ops = &ls1x_clk_divider_ops, \
> > + .parent_hws = (const struct clk_hw *[]) { _pname }, \
> > + .num_parents = 1, \
> > + .flags = CLK_GET_RATE_NOCACHE, \
> > + }, \
> > +}
> > +
> > +static LS1X_CLK_PLL(ls1b_clk_pll, CLK_PLL_FREQ, 12, 1, 0, 5, 0, 0);
> > +static LS1X_CLK_DIV(ls1b_clk_cpu, &ls1b_clk_pll.hw, CLK_PLL_DIV,
> > + 20, 4, NULL, 8, 0,
> > + CLK_DIVIDER_ONE_BASED | CLK_DIVIDER_ROUND_CLOSEST);
> > +static LS1X_CLK_DIV(ls1b_clk_dc, &ls1b_clk_pll.hw, CLK_PLL_DIV,
> > + 26, 4, NULL, 12, 0, CLK_DIVIDER_ONE_BASED);
> > +static LS1X_CLK_DIV(ls1b_clk_ahb, &ls1b_clk_pll.hw, CLK_PLL_DIV,
> > + 14, 4, NULL, 10, 0, CLK_DIVIDER_ONE_BASED);
> > +static CLK_FIXED_FACTOR(ls1b_clk_apb, "ls1b_clk_apb", "ls1b_clk_ahb", 2, 1,
> > + CLK_SET_RATE_PARENT);
> > +
> > +static struct clk_hw_onecell_data ls1b_clk_hw_data = {
> > + .hws = {
> > + [LS1X_CLKID_PLL] = &ls1b_clk_pll.hw,
> > + [LS1X_CLKID_CPU] = &ls1b_clk_cpu.hw,
> > + [LS1X_CLKID_DC] = &ls1b_clk_dc.hw,
> > + [LS1X_CLKID_AHB] = &ls1b_clk_ahb.hw,
> > + [LS1X_CLKID_APB] = &ls1b_clk_apb.hw,
> > + [CLK_NR_CLKS] = NULL,
>
> Do you need a CLK_NR_CLKS sentinel entry?
>
Not necessary.
Will remove.
> > + },
> > + .num = CLK_NR_CLKS,
> > +};
> > +
> > +static const struct clk_div_table ls1c_ahb_div_table[] = {
> > + [0] = { .val = 0, .div = 2 },
> > + [1] = { .val = 1, .div = 4 },
> > + [2] = { .val = 2, .div = 3 },
> > + [3] = { .val = 3, .div = 3 },
> > + [4] = { /* sentinel */ }
> > +};
> > +
> > +static LS1X_CLK_PLL(ls1c_clk_pll, CLK_PLL_FREQ, 0, 2, 8, 8, 16, 8);
> > +static LS1X_CLK_DIV(ls1c_clk_cpu, &ls1c_clk_pll.hw, CLK_PLL_DIV,
> > + 8, 7, NULL, 0, 1,
> > + CLK_DIVIDER_ONE_BASED | CLK_DIVIDER_ROUND_CLOSEST);
> > +static LS1X_CLK_DIV(ls1c_clk_dc, &ls1c_clk_pll.hw, CLK_PLL_DIV,
> > + 24, 7, NULL, 4, 1, CLK_DIVIDER_ONE_BASED);
> > +static LS1X_CLK_DIV(ls1c_clk_ahb, &ls1c_clk_cpu.hw, CLK_PLL_FREQ,
> > + 0, 2, ls1c_ahb_div_table, 0, 0, CLK_DIVIDER_ALLOW_ZERO);
> > +static CLK_FIXED_FACTOR(ls1c_clk_apb, "ls1c_clk_apb", "ls1c_clk_ahb", 1, 1,
> > + CLK_SET_RATE_PARENT);
> > +
> > +static struct clk_hw_onecell_data ls1c_clk_hw_data = {
> > + .hws = {
> > + [LS1X_CLKID_PLL] = &ls1c_clk_pll.hw,
> > + [LS1X_CLKID_CPU] = &ls1c_clk_cpu.hw,
> > + [LS1X_CLKID_DC] = &ls1c_clk_dc.hw,
> > + [LS1X_CLKID_AHB] = &ls1c_clk_ahb.hw,
> > + [LS1X_CLKID_APB] = &ls1c_clk_apb.hw,
> > + [CLK_NR_CLKS] = NULL,
> > + },
> > + .num = CLK_NR_CLKS,
> > +};
> > +
> > +static void __init ls1x_clk_init(struct device_node *np,
> > + struct clk_hw_onecell_data *hw_data)
> > +{
> > + struct ls1x_clk *ls1x_clk;
> > + void __iomem *reg;
> > + int i, ret;
> > +
> > + reg = of_iomap(np, 0);
> > + if (!reg) {
> > + pr_err("Unable to map base for %pOF\n", np);
>
> Needs include.
>
Will do.
> > + return;
> > + }
> > +
> > + for (i = 0; i < CLK_NR_CLKS; i++) {
> > + /* array might be sparse */
> > + if (!hw_data->hws[i])
> > + continue;
> > +
> > + if (i != LS1X_CLKID_APB) {
> > + ls1x_clk = to_ls1x_clk(hw_data->hws[i]);
> > + ls1x_clk->reg = reg + ls1x_clk->offset;
> > + }
> > +
> > + ret = of_clk_hw_register(np, hw_data->hws[i]);
> > + if (ret)
> > + return;
>
> unmap memory on failure? and unregister clks?
>
Will add the error handling.
> > + }
> > +
> > + ret = of_clk_add_hw_provider(np, of_clk_hw_onecell_get, hw_data);
> > + if (ret)
> > + pr_err("Failed to register %pOF\n", np);
>
> unmap memory on failure? And unregister clks?
>
Will add the error handling.
> > +}
> > +
> > +static void __init ls1b_clk_init(struct device_node *np)
> > +{
> > + return ls1x_clk_init(np, &ls1b_clk_hw_data);
> > +}
> > +
> > +static void __init ls1c_clk_init(struct device_node *np)
> > +{
> > + return ls1x_clk_init(np, &ls1c_clk_hw_data);
> > +}
> > +
> > +CLK_OF_DECLARE(ls1b_clk, "loongson,ls1b-clk", ls1b_clk_init);
> > +CLK_OF_DECLARE(ls1c_clk, "loongson,ls1c-clk", ls1c_clk_init);
>
> Any reason these can't be platform device drivers?
>
The Loongson1 system uses performance counter as the default clocksource.
Then, mips_hpt_frequency should be figured out when doing plat_time_init(),
just before calling mips_clockevent_init() and init_mips_clocksource().
And mips_hpt_frequency depends on CPU clock.
So these clocks should be initialized as early as possible.
It's too late for being a platform device driver.
> > +
> > +MODULE_AUTHOR("Keguang Zhang <keguang.zhang@xxxxxxxxx>");
> > +MODULE_DESCRIPTION("Loongson1 clock driver");
>
> It's not a module. So these are useless macros. Drop them?
Yes. I realized these macros will never be called.
Will drop.
--
Best regards,
Keguang Zhang
