[PATCH V2 5/8] clk: imx: add i.MX93 clk gate
Abel Vesa
abel.vesa at linaro.org
Mon Aug 15 02:07:51 PDT 2022
On 22-08-15 09:30:36, Peng Fan (OSS) wrote:
> From: Peng Fan <peng.fan at nxp.com>
>
> i.MX93 LPCG is different from i.MX8M CCGR. Although imx_clk_hw_gate4_flags
> is used here, it not strictly match i.MX93. i.MX93 has such design:
> - LPCG_DIRECT use BIT0 as on/off gate when LPCG_AUTHEN CPU_LPM is 0
> - LPCG_LPM_CUR use BIT[2:0] as on/off gate when LPCG_AUTHEN CPU_LPM is 1
>
> The current implementation suppose CPU_LPM is 0, and use LPCG_DIRECT
> BIT[1:0] as on/off gate. Although BIT1 is touched, actually BIT1 is
> reserved.
>
> And imx_clk_hw_gate4_flags use mask 0x3 to determine whether the clk
> is enabled or not, but i.MX93 LPCG only use BIT0 to control when CPU_LPM
> is 0. So clk disabled unused during kernel boot not able to gate off
> the unused clocks.
>
> To match i.MX93 LPCG, introduce imx93_clk_gate.
>
> Signed-off-by: Peng Fan <peng.fan at nxp.com>
> Reviewed-by: Ye Li <ye.li at nxp.com>
> Reviewed-by: Jacky Bai <ping.bai at nxp.com>
> ---
> drivers/clk/imx/Makefile | 2 +-
> drivers/clk/imx/clk-gate-93.c | 199 ++++++++++++++++++++++++++++++++++
> drivers/clk/imx/clk.h | 4 +
> 3 files changed, 204 insertions(+), 1 deletion(-)
> create mode 100644 drivers/clk/imx/clk-gate-93.c
>
> diff --git a/drivers/clk/imx/Makefile b/drivers/clk/imx/Makefile
> index 88b9b9285d22..89fe72327788 100644
> --- a/drivers/clk/imx/Makefile
> +++ b/drivers/clk/imx/Makefile
> @@ -28,7 +28,7 @@ obj-$(CONFIG_CLK_IMX8MN) += clk-imx8mn.o
> obj-$(CONFIG_CLK_IMX8MP) += clk-imx8mp.o
> obj-$(CONFIG_CLK_IMX8MQ) += clk-imx8mq.o
>
> -obj-$(CONFIG_CLK_IMX93) += clk-imx93.o
> +obj-$(CONFIG_CLK_IMX93) += clk-imx93.o clk-gate-93.o
>
> obj-$(CONFIG_MXC_CLK_SCU) += clk-imx-scu.o clk-imx-lpcg-scu.o
> clk-imx-scu-$(CONFIG_CLK_IMX8QXP) += clk-scu.o clk-imx8qxp.o \
> diff --git a/drivers/clk/imx/clk-gate-93.c b/drivers/clk/imx/clk-gate-93.c
> new file mode 100644
> index 000000000000..ceb56b290394
> --- /dev/null
> +++ b/drivers/clk/imx/clk-gate-93.c
> @@ -0,0 +1,199 @@
> +// SPDX-License-Identifier: GPL-2.0+
> +/*
> + * Copyright 2022 NXP
> + *
> + * Peng Fan <peng.fan at nxp.com>
> + */
> +
> +#include <linux/clk-provider.h>
> +#include <linux/errno.h>
> +#include <linux/export.h>
> +#include <linux/io.h>
> +#include <linux/iopoll.h>
> +#include <linux/slab.h>
> +
> +#include "clk.h"
> +
> +#define DIRECT_OFFSET 0x0
> +
> +/*
> + * 0b000 - LPCG will be OFF in any CPU mode.
> + * 0b100 - LPCG will be ON in any CPU mode.
> + */
> +#define LPM_SETTING_OFF 0x0
> +#define LPM_SETTING_ON 0x4
> +
> +#define LPM_CUR_OFFSET 0x1c
> +
> +#define AUTHEN_OFFSET 0x30
> +#define CPULPM_EN BIT(2)
> +#define TZ_NS_SHIFT 9
> +#define TZ_NS_MASK BIT(9)
> +
> +#define WHITE_LIST_SHIFT 16
> +
> +struct imx93_clk_gate {
> + struct clk_hw hw;
> + void __iomem *reg;
> + u32 bit_idx;
> + u32 val;
> + u32 mask;
> + spinlock_t *lock;
> + unsigned int *share_count;
> +};
> +
> +#define to_imx93_clk_gate(_hw) container_of(_hw, struct imx93_clk_gate, hw)
> +
> +static void imx93_clk_gate_do_hardware(struct clk_hw *hw, bool enable)
> +{
> + struct imx93_clk_gate *gate = to_imx93_clk_gate(hw);
> + u32 val;
> +
> + val = readl(gate->reg + AUTHEN_OFFSET);
> + if (val & CPULPM_EN) {
> + val = enable ? LPM_SETTING_ON : LPM_SETTING_OFF;
> + writel(val, gate->reg + LPM_CUR_OFFSET);
> + } else {
> + val = readl(gate->reg + DIRECT_OFFSET);
> + val &= ~(gate->mask << gate->bit_idx);
> + if (enable)
> + val |= (gate->val & gate->mask) << gate->bit_idx;
> + writel(val, gate->reg + DIRECT_OFFSET);
> + }
> +}
> +
> +static int imx93_clk_gate_enable(struct clk_hw *hw)
> +{
> + struct imx93_clk_gate *gate = to_imx93_clk_gate(hw);
> + unsigned long flags;
> +
> + spin_lock_irqsave(gate->lock, flags);
> +
> + if (gate->share_count && (*gate->share_count)++ > 0)
> + goto out;
> +
> + imx93_clk_gate_do_hardware(hw, true);
> +out:
> + spin_unlock_irqrestore(gate->lock, flags);
> +
> + return 0;
> +}
Just wondering if we could use the existing clk-gate2 since we would
only have to implement the ops that are different there.
Also, would the next i.MX9 platforms also use this? Or will we have one
similar driver for each new platform?
> +
> +static void imx93_clk_gate_disable(struct clk_hw *hw)
> +{
> + struct imx93_clk_gate *gate = to_imx93_clk_gate(hw);
> + unsigned long flags;
> +
> + spin_lock_irqsave(gate->lock, flags);
> +
> + if (gate->share_count) {
> + if (WARN_ON(*gate->share_count == 0))
> + goto out;
> + else if (--(*gate->share_count) > 0)
> + goto out;
> + }
> +
> + imx93_clk_gate_do_hardware(hw, false);
> +out:
> + spin_unlock_irqrestore(gate->lock, flags);
> +}
> +
> +static int imx93_clk_gate_reg_is_enabled(struct imx93_clk_gate *gate)
> +{
> + u32 val = readl(gate->reg + AUTHEN_OFFSET);
> +
> + if (val & CPULPM_EN) {
> + val = readl(gate->reg + LPM_CUR_OFFSET);
> + if (val == LPM_SETTING_ON)
> + return 1;
> + } else {
> + val = readl(gate->reg);
> + if (((val >> gate->bit_idx) & gate->mask) == gate->val)
> + return 1;
> + }
> +
> + return 0;
> +}
> +
> +static int imx93_clk_gate_is_enabled(struct clk_hw *hw)
> +{
> + struct imx93_clk_gate *gate = to_imx93_clk_gate(hw);
> + unsigned long flags;
> + int ret;
> +
> + spin_lock_irqsave(gate->lock, flags);
> +
> + ret = imx93_clk_gate_reg_is_enabled(gate);
> +
> + spin_unlock_irqrestore(gate->lock, flags);
> +
> + return ret;
> +}
> +
> +static void imx93_clk_gate_disable_unused(struct clk_hw *hw)
> +{
> + struct imx93_clk_gate *gate = to_imx93_clk_gate(hw);
> + unsigned long flags;
> +
> + spin_lock_irqsave(gate->lock, flags);
> +
> + if (!gate->share_count || *gate->share_count == 0)
> + imx93_clk_gate_do_hardware(hw, false);
> +
> + spin_unlock_irqrestore(gate->lock, flags);
> +}
> +
> +static const struct clk_ops imx93_clk_gate_ops = {
> + .enable = imx93_clk_gate_enable,
> + .disable = imx93_clk_gate_disable,
> + .disable_unused = imx93_clk_gate_disable_unused,
> + .is_enabled = imx93_clk_gate_is_enabled,
> +};
> +
> +static const struct clk_ops imx93_clk_gate_ro_ops = {
> + .is_enabled = imx93_clk_gate_is_enabled,
> +};
> +
> +struct clk_hw *imx93_clk_gate(struct device *dev, const char *name, const char *parent_name,
> + unsigned long flags, void __iomem *reg, u32 bit_idx, u32 val,
> + u32 mask, u32 domain_id, unsigned int *share_count)
> +{
> + struct imx93_clk_gate *gate;
> + struct clk_hw *hw;
> + struct clk_init_data init;
> + int ret;
> + u32 authen;
> +
> + gate = kzalloc(sizeof(struct imx93_clk_gate), GFP_KERNEL);
> + if (!gate)
> + return ERR_PTR(-ENOMEM);
> +
> + gate->reg = reg;
> + gate->lock = &imx_ccm_lock;
> + gate->bit_idx = bit_idx;
> + gate->val = val;
> + gate->mask = mask;
> + gate->share_count = share_count;
> +
> + init.name = name;
> + init.ops = &imx93_clk_gate_ops;
> + init.flags = flags | CLK_SET_RATE_PARENT | CLK_OPS_PARENT_ENABLE;
> + init.parent_names = parent_name ? &parent_name : NULL;
> + init.num_parents = parent_name ? 1 : 0;
> +
> + gate->hw.init = &init;
> + hw = &gate->hw;
> +
> + authen = readl(reg + AUTHEN_OFFSET);
> + if (!(authen & TZ_NS_MASK) || !(authen & BIT(WHITE_LIST_SHIFT + domain_id)))
> + init.ops = &imx93_clk_gate_ro_ops;
> +
> + ret = clk_hw_register(dev, hw);
> + if (ret) {
> + kfree(gate);
> + return ERR_PTR(ret);
> + }
> +
> + return hw;
> +}
> +EXPORT_SYMBOL_GPL(imx93_clk_gate);
> diff --git a/drivers/clk/imx/clk.h b/drivers/clk/imx/clk.h
> index 396a5ea75083..dd49f90110e8 100644
> --- a/drivers/clk/imx/clk.h
> +++ b/drivers/clk/imx/clk.h
> @@ -451,6 +451,10 @@ struct clk_hw *imx93_clk_composite_flags(const char *name,
> imx93_clk_composite_flags(name, parent_names, num_parents, reg, domain_id \
> CLK_SET_RATE_NO_REPARENT | CLK_OPS_PARENT_ENABLE)
>
> +struct clk_hw *imx93_clk_gate(struct device *dev, const char *name, const char *parent_name,
> + unsigned long flags, void __iomem *reg, u32 bit_idx, u32 val,
> + u32 mask, u32 domain_id, unsigned int *share_count);
> +
> struct clk_hw *imx_clk_hw_divider_gate(const char *name, const char *parent_name,
> unsigned long flags, void __iomem *reg, u8 shift, u8 width,
> u8 clk_divider_flags, const struct clk_div_table *table,
> --
> 2.37.1
>
More information about the linux-arm-kernel
mailing list