Quoting jiada_wang@xxxxxxxxxx (2018-10-25 00:23:46)
> diff --git a/drivers/clk/renesas/Makefile b/drivers/clk/renesas/Makefile
> index 71d4cafe15c0..17b05955e4f4 100644
> --- a/drivers/clk/renesas/Makefile
> +++ b/drivers/clk/renesas/Makefile
> @@ -34,3 +34,4 @@ obj-$(CONFIG_CLK_RCAR_USB2_CLOCK_SEL) += rcar-usb2-clock-sel.o
> obj-$(CONFIG_CLK_RENESAS_CPG_MSSR) += renesas-cpg-mssr.o
> obj-$(CONFIG_CLK_RENESAS_CPG_MSTP) += clk-mstp.o
> obj-$(CONFIG_CLK_RENESAS_DIV6) += clk-div6.o
> +obj-$(CONFIG_CLK_RENESAS_CLK_AVB) += clk-avb.o
Any reason this can't be placed in some sort of alphabetical order
instead of at the end of the file?
> diff --git a/drivers/clk/renesas/clk-avb.c b/drivers/clk/renesas/clk-avb.c
> new file mode 100644
> index 000000000000..bb1eef0e9bee
> --- /dev/null
> +++ b/drivers/clk/renesas/clk-avb.c
> @@ -0,0 +1,257 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2017 Mentor
> + *
> + * Contact: Jiada Wang <jiada_wang@xxxxxxxxxx>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; version 2 of the License.
Can you remove this paragraph? It duplicates the SPDX tag.
> + *
> + * avb Common Clock Framework support
> + */
> +
> +#include <linux/clk-provider.h>
> +#include <linux/of_address.h>
> +#include <linux/of_platform.h>
> +#include <linux/slab.h>
> +
> +struct clk_avb_data {
> + void __iomem *base;
> +
> + struct clk_onecell_data clk_data;
> + /* lock reg access */
> + spinlock_t lock;
> +};
> +
> +struct clk_avb {
> + struct clk_hw hw;
> + unsigned int idx;
> + struct clk_avb_data *data;
> +};
> +
> +#define to_clk_avb(_hw) container_of(_hw, struct clk_avb, hw)
> +
> +#define AVB_DIV_MASK 0x3ffff
> +#define AVB_MAX_DIV 0x3ffc0
> +#define AVB_COUNTER_MAX_FREQ 25000000
> +#define AVB_COUNTER_NUM 8
> +#define AVB_CLK_NAME_SIZE 10
> +#define AVB_ID_TO_DIV(id) ((id) * 4)
> +
> +#define AVB_CLK_CONFIG 0x20
> +#define AVB_DIV_EN_COM BIT(31)
> +#define AVB_CLK_NAME "avb"
> +#define ADG_CLK_NAME "adg"
> +
> +static int clk_avb_is_enabled(struct clk_hw *hw)
> +{
> + struct clk_avb *avb = to_clk_avb(hw);
> +
> + return (clk_readl(avb->data->base + AVB_CLK_CONFIG) & BIT(avb->idx));
Please drop the extra parenthesis.
> +}
> +
> +static int clk_avb_enabledisable(struct clk_hw *hw, int enable)
> +{
> + struct clk_avb *avb = to_clk_avb(hw);
> + u32 val;
> +
> + spin_lock(&avb->data->lock);
> +
> + val = clk_readl(avb->data->base + AVB_CLK_CONFIG);
> + if (enable)
> + val |= BIT(avb->idx);
> + else
> + val &= ~BIT(avb->idx);
> + clk_writel(val, avb->data->base + AVB_CLK_CONFIG);
> +
> + spin_unlock(&avb->data->lock);
> +
> + return 0;
> +}
> +
> +static int clk_avb_enable(struct clk_hw *hw)
> +{
> + return clk_avb_enabledisable(hw, 1);
> +}
> +
> +static void clk_avb_disable(struct clk_hw *hw)
> +{
> + clk_avb_enabledisable(hw, 0);
> +}
> +
> +static unsigned long clk_avb_recalc_rate(struct clk_hw *hw,
> + unsigned long parent_rate)
> +{
> + struct clk_avb *avb = to_clk_avb(hw);
> + u32 div;
> +
> + div = clk_readl(avb->data->base + AVB_ID_TO_DIV(avb->idx)) &
> + AVB_DIV_MASK;
Split this to two lines:
div = readl(...);
div &= AVB_DIV_MASK;
> + if (!div)
> + return parent_rate;
> +
> + return parent_rate * 32 / div;
> +}
> +
> +static unsigned int clk_avb_calc_div(unsigned long rate,
> + unsigned long parent_rate)
> +{
> + unsigned int div;
> +
> + if (!rate)
> + rate = 1;
> +
> + if (rate > AVB_COUNTER_MAX_FREQ)
> + rate = AVB_COUNTER_MAX_FREQ;
rate = min(rate, AVB_COUNTER_MAX_FREQ);
> +
> + div = DIV_ROUND_CLOSEST(parent_rate * 32, rate);
> +
> + if (div > AVB_MAX_DIV)
> + div = AVB_MAX_DIV;
div = min(div, AVB_MAX_DIV);
> +
> + return div;
> +}
> +
> +static long clk_avb_round_rate(struct clk_hw *hw, unsigned long rate,
> + unsigned long *parent_rate)
> +{
> + unsigned int div = clk_avb_calc_div(rate, *parent_rate);
> +
> + return (*parent_rate * 32) / div;
> +}
> +
> +static int clk_avb_set_rate(struct clk_hw *hw, unsigned long rate,
> + unsigned long parent_rate)
> +{
> + struct clk_avb *avb = to_clk_avb(hw);
> + unsigned int div = clk_avb_calc_div(rate, parent_rate);
> + u32 val;
> +
> + val = clk_readl(avb->data->base + AVB_ID_TO_DIV(avb->idx)) &
> + ~AVB_DIV_MASK;
> + clk_writel(val | div, avb->data->base + AVB_ID_TO_DIV(avb->idx));
Any reason to use clk_readl/writel? Preferably these APIs aren't used
and you just use readl/writel instead.
> +
> + return 0;
> +}
> +
> +static const struct clk_ops clk_avb_ops = {
> + .enable = clk_avb_enable,
> + .disable = clk_avb_disable,
> + .is_enabled = clk_avb_is_enabled,
> + .recalc_rate = clk_avb_recalc_rate,
> + .round_rate = clk_avb_round_rate,
> + .set_rate = clk_avb_set_rate,
> +};
> +
> +static struct clk *clk_register_avb(struct clk_avb_data *data,
> + unsigned int id)
> +{
> + struct clk_init_data init;
> + struct clk_avb *avb;
> + struct clk *clk;
> + char name[AVB_CLK_NAME_SIZE];
> + const char *parent_name = ADG_CLK_NAME;
> +
> + avb = kzalloc(sizeof(*avb), GFP_KERNEL);
> + if (!avb)
> + return ERR_PTR(-ENOMEM);
> +
> + snprintf(name, AVB_CLK_NAME_SIZE, "%s.%u", AVB_CLK_NAME, id);
> +
> + avb->idx = id;
> + avb->data = data;
> +
> + /* Register the clock. */
> + init.name = name;
> + init.ops = &clk_avb_ops;
> + init.flags = CLK_IS_BASIC;
> + init.parent_names = &parent_name;
> + init.num_parents = 1;
> +
> + avb->hw.init = &init;
> +
> + /* init DIV to a valid state */
> + writel(AVB_MAX_DIV, data->base + AVB_ID_TO_DIV(avb->idx));
> +
> + clk = clk_register(NULL, &avb->hw);
> + if (IS_ERR(clk))
> + kfree(avb);
> +
> + return clk;
> +}
> +
> +static void clk_unregister_avb(struct clk *clk)
> +{
> + struct clk_avb *avb;
> + struct clk_hw *hw;
> +
> + if (IS_ERR(clk))
> + return;
> +
> + hw = __clk_get_hw(clk);
> + if (!hw)
> + return;
> +
> + avb = to_clk_avb(hw);
> +
> + clk_unregister(clk);
> + kfree(avb);
Using clk_hw based registration APIs would make this a little cleaner.
> +}
> +
> +static void __init clk_avb_setup(struct device_node *node)
> +{
> + struct clk_avb_data *data;
> + struct clk_onecell_data *clk_data;
> + int ret, i;
> + struct resource res;
> +
> + data = kzalloc(sizeof(*data), GFP_KERNEL);
> + if (!data)
> + return;
> +
> + data->base = of_io_request_and_map(node, 0, of_node_full_name(node));
> + if (IS_ERR(data->base))
> + goto err_data;
> +
> + spin_lock_init(&data->lock);
> +
> + clk_data = &data->clk_data;
> + clk_data->clk_num = AVB_COUNTER_NUM;
> + clk_data->clks = kmalloc_array(AVB_COUNTER_NUM,
> + sizeof(struct clk *),
> + GFP_KERNEL);
> + if (!clk_data->clks)
> + goto err_unmap;
> +
> + for (i = 0; i < AVB_COUNTER_NUM; i++) {
> + clk_data->clks[i] = clk_register_avb(data, i);
> + if (IS_ERR(clk_data->clks[i])) {
> + pr_err("failed to register clock %s.%d\n",
> + AVB_CLK_NAME, i);
> + goto err_clks;
> + }
> + }
> +
> + ret = of_clk_add_provider(node, of_clk_src_onecell_get, clk_data);
Can you use the clk_hw based registration and provider APIs?
> + if (ret) {
> + pr_err("failed to register clock provider\n");
> + goto err_clks;
> + }
> +
> + writel(AVB_DIV_EN_COM, data->base + AVB_CLK_CONFIG);
> +
> + return;
> +
> +err_clks:
> + for (i = 0; i < AVB_COUNTER_NUM; i++)
> + clk_unregister_avb(clk_data->clks[i]);
> +err_unmap:
> + iounmap(data->base);
> + of_address_to_resource(node, 0, &res);
> + release_mem_region(res.start, resource_size(&res));
> +err_data:
> + kfree(data);
> +}
> +
> +CLK_OF_DECLARE(avb, "renesas,clk-avb", clk_avb_setup);
Any reason to not use platform device APIs? If not, please change this
to a platform driver. Otherwise, add a comment indicating why it can't
be done.
