Re: [PATCH v5 04/14] clk: clk-pic32: Add PIC32 clock driver

From: Purna Chandra Mandal
Date: Wed Feb 03 2016 - 00:37:45 EST


On 01/30/2016 05:28 AM, Stephen Boyd wrote:

> On 01/13, Joshua Henderson wrote:
>> diff --git a/drivers/clk/clk-pic32.c b/drivers/clk/clk-pic32.c
>> new file mode 100644
>> index 0000000..9dc5f78
>> --- /dev/null
>> +++ b/drivers/clk/clk-pic32.c
>> @@ -0,0 +1,1801 @@
>> +/*
>> + * Purna Chandra Mandal,<purna.mandal@xxxxxxxxxxxxx>
>> + * Copyright (C) 2015 Microchip Technology Inc. All rights reserved.
>> + *
>> + * This program is free software; you can distribute it and/or modify it
>> + * under the terms of the GNU General Public License (Version 2) as
>> + * published by the Free Software Foundation.
>> + *
>> + * This program is distributed in the hope it will be useful, but WITHOUT
>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
>> + * for more details.
>> + */
>> +#include <linux/interrupt.h>
>> +#include <linux/clk.h>
> Is this used?

Yes. used below. Will remove as discussed below.

>> +#include <linux/clkdev.h>
>> +#include <linux/clk-provider.h>
>> +#include <linux/of_address.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/of.h>
>> +#include <linux/of_irq.h>
>> +#include <linux/delay.h>
>> +#include <linux/debugfs.h>
>> +#include <asm/traps.h>
>> +#include <linux/slab.h>
> Please move asm includes after all linux includes.
>
> Include <linux/kernel.h> for abs usage?
>
ack.

>> +
>> +#include <asm/mach-pic32/pic32.h>
> Is this header required? I'd like to be able to build this file
> without using a MIPS compiler. For example, if we could move that
> pic32_syskey_unlock() prototype somewhere else besides
> asm/mach-pic32 then this should compile fine on non-MIPS kernels?

Yes it is good to remove the dependency for compilation purpose. But this
driver uses few more functionality/macros (like PIC32_CLR, PIC32_SET)
defined in pic32.h which are used by this driver to select
atomic-register-offset.

>> +
>> +/* OSCCON Reg fields */
>> +#define OSC_CUR_MASK 0x07
>> +#define OSC_CUR_SHIFT 12
>> +#define OSC_NEW_MASK 0x07
>> +#define OSC_NEW_SHIFT 8
>> +#define OSC_SWEN 0x01
>> +#define OSC_CLK_FAILED 0x04
>> +
>> +/* SPLLCON Reg fields */
>> +#define PLL_RANGE_MASK 0x07
>> +#define PLL_RANGE_SHIFT 0
>> +#define PLL_ICLK_MASK 0x01
>> +#define PLL_ICLK_SHIFT 7
>> +#define PLL_IDIV_MASK 0x07
>> +#define PLL_IDIV_SHIFT 8
>> +#define PLL_ODIV_MASK 0x07
>> +#define PLL_ODIV_SHIFT 24
>> +#define PLL_MULT_MASK 0x7F
>> +#define PLL_MULT_SHIFT 16
>> +#define PLL_MULT_MAX 128
>> +#define PLL_ODIV_MIN 1
>> +#define PLL_ODIV_MAX 5
>> +
>> +/* Peripheral Bus Clock Reg Fields */
>> +#define PB_DIV_MASK 0x7f
>> +#define PB_DIV_SHIFT 0
>> +#define PB_DIV_READY BIT(11)
>> +#define PB_DIV_ENABLED BIT(15)
>> +#define PB_DIV_MAX 128
>> +#define PB_DIV_MIN 0
>> +
>> +/* Reference Oscillator Control Reg fields */
>> +#define REFO_SEL_MASK 0x0f
>> +#define REFO_SEL_SHIFT 0
>> +#define REFO_ACTIVE BIT(8)
>> +#define REFO_DIVSW_EN BIT(9)
>> +#define REFO_OE BIT(12)
>> +#define REFO_ON BIT(15)
>> +#define REFO_DIV_SHIFT 16
>> +#define REFO_DIV_MASK 0x7fff
>> +
>> +/* Reference Oscillator Trim Register Fields */
>> +#define REFO_TRIM_REG 0x10 /* Register offset w.r.t. REFO_CON_REG */
>> +#define REFO_TRIM_MASK 0x1ff
>> +#define REFO_TRIM_SHIFT 23
>> +#define REFO_TRIM_MAX 511
>> +
>> +/* FRC postscaler */
>> +#define OSC_FRCDIV_MASK 0x07
>> +#define OSC_FRCDIV_SHIFT 24
>> +
>> +/* FRC tuning */
>> +#define OSC_FRCTUN_MASK 0x3F
>> +#define OSC_FRCTUN_SHIFT 0
>> +
>> +/* SLEW Control Register fields */
>> +#define SLEW_BUSY 0x01
>> +#define SLEW_DOWNEN 0x02
>> +#define SLEW_UPEN 0x04
>> +#define SLEW_DIV 0x07
>> +#define SLEW_DIV_SHIFT 8
>> +#define SLEW_SYSDIV 0x0f
>> +#define SLEW_SYSDIV_SHIFT 20
>> +
>> +/* Common clock flags */
>> +#define CLK_ENABLED_ALWAYS CLK_IGNORE_UNUSED
>> +#define CLK_DIV_FIXED BIT(20)
>> +
>> +/* Sys Mux clock flags */
>> +#define SYS_MUX_POSTDIV 0x1
>> +#define SYS_MUX_SLEW 0x2
>> +
>> +#define LOCK_TIMEOUT_NS (100 * NSEC_PER_MSEC)
>> +
>> +/* System PLL clk */
>> +struct pic32_spll {
>> + struct clk_hw hw;
>> + void __iomem *regs;
>> + void __iomem *status_reg;
>> + u32 pll_locked;
> Maybe this could be called lock_mask?

ack.

>> + u8 idiv; /* pll-iclk divider, treating fixed */
>> +};
>> +
>> +/* System Clk */
>> +struct pic32_sclk {
>> + struct clk_hw hw;
>> + void __iomem *regs;
>> + void __iomem *slwreg;
>> + unsigned long flags;
>> + u32 *parent_idx;
> #ifdef CONFIG_DEBUGFS?

ack. Will remove it and corresponding clk operation callback 'debug_init()' as well.

>> + struct debugfs_regset32 regset;
>> +};
>> +
>> +/* Reference Oscillator */
>> +struct pic32_refosc {
>> + struct clk_hw hw;
>> + void __iomem *regs;
>> + u32 *parent_idx;
>> + struct debugfs_regset32 regset;
>> +};
>> +
>> +/* Peripheral Bus Clock */
>> +struct pic32_pbclk {
>> + struct clk_hw hw;
>> + void __iomem *regs;
>> + u32 flags;
> This should probably be unsigned long.

ack.

>> + struct debugfs_regset32 regset;
>> +};
>> +
>> +/* External SOSC(fixed gated) clock */
>> +struct pic32_sosc {
>> + struct clk_hw hw;
>> + void __iomem *regs;
>> + void __iomem *status_reg;
>> + unsigned long fixed_rate;
>> + int bitmask;
>> + int status_bitmask;
>> +};
>> +
>> +/* Soc specific clock reg-base */
>> +static void __iomem *pic32_clk_regbase;
>> +static struct clk *pic32_sys_clk;
>> +
>> +static DEFINE_SPINLOCK(lock);
>> +
>> +#define __clk_lock(flags) spin_lock_irqsave(&lock, flags)
>> +#define __clk_unlock(flags) spin_unlock_irqrestore(&lock, flags)
>> +
>> +/* execute unlock-sequence before writing to system registers */
>> +#define pic32_devcon_sysunlock() pic32_syskey_unlock()
>> +#define pic32_devcon_syslock()
> Will this be defined to something one day?

Yes, in future it might be used in production software.
Will remove it now.

>> +
>> +/* add instruction pipeline delay while CPU clock is in-transition. */
>> +#define cpu_nop5() \
>> +do { \
>> + __asm__ __volatile__("nop"); \
>> + __asm__ __volatile__("nop"); \
>> + __asm__ __volatile__("nop"); \
>> + __asm__ __volatile__("nop"); \
>> + __asm__ __volatile__("nop"); \
>> +} while (0)
>> +
>> +#define clkhw_to_spll(_hw) container_of(_hw, struct pic32_spll, hw)
>> +#define clkhw_to_refosc(_hw) container_of(_hw, struct pic32_refosc, hw)
>> +#define clkhw_to_pbclk(_hw) container_of(_hw, struct pic32_pbclk, hw)
>> +#define clkhw_to_sys_clk(_hw) container_of(_hw, struct pic32_sclk, hw)
>> +#define clkhw_to_sosc(_hw) container_of(_hw, struct pic32_sosc, hw)
>> +
>> +/* pic32_of_clk_get_parent_indices - get parent clk hardware indices.
>> + *
>> + * This is useful specifically for mux clocks where some of possible parent-
>> + * clocks logically been dropped thereby creating discontinuous linear
>> + * sequence. This API refers OF property "microchip,clock-indices" of the
>> + * device node to find h/w id(s) corresponding to each input clock source.
>> + */
> This is not kernel doc notation.

ack. Will update as per kernel doc notation.

>> +int pic32_of_clk_get_parent_indices(struct device_node *np,
>> + u32 **table_p,
>> + int count)
>> +{
>> + struct property *prop;
>> + const __be32 *pv;
>> + u32 i, *array, ret;
>> +
>> + if ((!table_p) || (!count))
>> + return -EINVAL;
>> +
>> + prop = of_find_property(np, "microchip,clock-indices", NULL);
>> + if (!prop) {
>> + ret = 0;
>> + goto out_err;
>> + }
>> +
>> + array = kzalloc((sizeof(u32) * count), GFP_KERNEL);
> Who frees this? kcalloc too.

Good catch. Caller will have to free the memory as currently being done. Will add comment accordingly.

>> + if (!array) {
>> + ret = -ENOMEM;
>> + goto out_err;
>> + }
>> +
>> + for (i = 0, pv = NULL; i < count; i++) {
>> + pv = of_prop_next_u32(prop, pv, &array[i]);
>> + if (!pv) {
>> + kfree(array);
>> + ret = -EINVAL;
>> + goto out_err;
>> + }
>> + }
>> +
>> + *table_p = array;
>> + return 0;
>> +out_err:
>> + *table_p = NULL;
>> + return ret;
>> +}
>> +EXPORT_SYMBOL(pic32_of_clk_get_parent_indices);
> And who uses this API?

Some platform drivers (e.g. peripheral timer) uses this to select input clock
source(s). It might better be a platform API, but added here clock driver as it
is related to clock functionality and this driver uses it.

>> +
>> +static int pic32_of_clk_register_clkdev(struct device_node *np, struct clk *clk)
>> +{
>> + int ret;
>> +
>> + ret = clk_register_clkdev(clk, NULL, __clk_get_name(clk));
>> + if (ret) {
>> + pr_err("%s: clkdev register failed, ret %d\n",
>> + __clk_get_name(clk), ret);
>> + goto out_err;
>> + }
>> +
>> + ret = of_clk_add_provider(np, of_clk_src_simple_get, clk);
>> +
>> +out_err:
>> + return ret;
>> +}
>> +
>> +static int pbclk_endisable(struct clk_hw *hw, int enable)
>> +{
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> +
>> + if (enable)
>> + clk_writel(PB_DIV_ENABLED, PIC32_SET(pb->regs));
> Please don't use clk_writel. Just use readl/writel directly.

ack.

>> + else
>> + clk_writel(PB_DIV_ENABLED, PIC32_CLR(pb->regs));
>> + return 0;
>> +}
>> +
>> +static int pbclk_is_enabled(struct clk_hw *hw)
>> +{
>> + u32 v;
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> +
>> + v = clk_readl(pb->regs) & PB_DIV_ENABLED;
>> + return !!v;
>> +}
>> +
>> +static int pbclk_enable(struct clk_hw *hw)
>> +{
>> + return pbclk_endisable(hw, 1);
>> +}
>> +
>> +static void pbclk_disable(struct clk_hw *hw)
>> +{
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> +
>> + if (pb->flags & CLK_ENABLED_ALWAYS)
>> + return;
>> +
>> + pbclk_endisable(hw, 0);
>> + cpu_relax();
>> +}
>> +
>> +static unsigned long calc_best_divided_rate(unsigned long rate,
>> + unsigned long parent_rate,
>> + u32 divider_max,
>> + u32 divider_min)
>> +{
>> + u32 divided_rate_up, divided_rate_down, best_rate;
>> + u32 divider_down, divider_up;
>> +
>> + /* eq. clk_rate = parent_rate / divider.
>> + *
>> + * Find best divider to produce closest of target divided rate.
>> + */
>> +
>> + divider_down = parent_rate / rate;
>> + divider_up = divider_down + 1;
>> + if (divider_down >= divider_max) {
>> + divider_down = divider_max;
>> + divider_up = divider_down;
>> + } else if (divider_down < divider_min) {
>> + divider_down = divider_min;
>> + }
>> + divided_rate_up = parent_rate / divider_down;
>> + divided_rate_down = parent_rate / divider_up;
>> + if (abs(rate - divided_rate_down) < abs(rate - divided_rate_up))
>> + best_rate = divided_rate_down;
>> + else
>> + best_rate = divided_rate_up;
>> +
>> + return best_rate;
>> +}
>> +
>> +static inline u16 pbclk_read_pbdiv(struct pic32_pbclk *pb)
>> +{
>> + u32 v = clk_readl(pb->regs);
>> +
>> + return ((v >> PB_DIV_SHIFT) & PB_DIV_MASK) + 1;
>> +}
>> +
>> +static unsigned long pbclk_recalc_rate(struct clk_hw *hw,
>> + unsigned long parent_rate)
>> +{
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> + unsigned long div, rate;
>> +
>> + div = pbclk_read_pbdiv(pb);
>> + rate = parent_rate / div;
>> +
>> + return rate;
>> +}
>> +
>> +static long pbclk_round_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long *parent_rate)
>> +{
>> + long best_rate = calc_best_divided_rate(rate, *parent_rate,
>> + PB_DIV_MAX, PB_DIV_MIN);
>> + return best_rate;
>> +}
>> +
>> +static int pbclk_set_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long parent_rate)
>> +{
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> + u16 div, new_div;
>> + unsigned long pbclk, flags, v;
>> + ktime_t timeout;
>> +
>> + /* fixed-div clk ? */
>> + if (pb->flags & CLK_DIV_FIXED)
>> + return -EINVAL;
>> +
>> + /* calculate clkdiv and best rate */
>> + new_div = parent_rate / rate;
>> + pbclk = parent_rate / new_div;
>> +
>> + /* check & wait for PBDIVRDY */
>> + timeout = ktime_add_ns(ktime_get(), LOCK_TIMEOUT_NS);
>> + for (;;) {
>> + v = clk_readl(pb->regs);
>> + if (v & PB_DIV_READY)
>> + break;
>> +
>> + if (ktime_after(ktime_get(), timeout)) {
>> + pr_err("%s: pre_rate, busy while timeout\n",
>> + clk_hw_get_name(hw));
>> + return -EPERM;
>> + }
>> + cpu_relax();
>> + }
>> +
>> + __clk_lock(flags);
>> +
>> + /* apply new pbdiv */
>> + v = clk_readl(pb->regs);
>> + v &= ~PB_DIV_MASK;
>> + v |= (new_div - 1);
>> +
>> + /* sys unlock */
>> + pic32_devcon_sysunlock();
>> +
>> + clk_writel(v, pb->regs);
>> +
>> + /* sys lock */
>> + pic32_devcon_syslock();
>> +
>> + __clk_unlock(flags);
>> +
>> + /* wait again, for pbdivready */
>> + timeout = ktime_add_ns(ktime_get(), LOCK_TIMEOUT_NS);
>> + for (;;) {
>> + v = clk_readl(pb->regs);
>> + if (v & PB_DIV_READY)
>> + break;
>> +
>> + if (ktime_after(ktime_get(), timeout)) {
>> + pr_err("%s: post_rate, busy while timeout\n",
>> + clk_hw_get_name(hw));
>> + break;
>> + }
>> + }
> We have iopoll for this stuff too.

ack. Will replace with iopoll API.

>> +
>> + /* confirm that new div is applied correctly */
>> + div = pbclk_read_pbdiv(pb);
>> + return (div == new_div) ? 0 : -EPERM;
>> +}
>> +
>> +static struct debugfs_reg32 pbclk_regs_debug[] = {
> Can this be const? Same goes for all these debugfs_reg32 arrays.

ack. Will update.

>> + { .name = "PBxDIV", .offset = 0,},
>> +};
>> +
>> +static int pbclk_debug_init(struct clk_hw *hw, struct dentry *dentry)
>> +{
>> + struct pic32_pbclk *pb = clkhw_to_pbclk(hw);
>> + struct dentry *file;
>> +
>> + pb->regset.base = pb->regs;
>> + pb->regset.regs = pbclk_regs_debug;
>> + pb->regset.nregs = ARRAY_SIZE(pbclk_regs_debug);
>> +
>> + file = debugfs_create_regset32("regdump", S_IRUGO, dentry, &pb->regset);
>> + if (IS_ERR(file))
>> + return PTR_ERR(file);
>> +
>> + return 0;
>> +}
>> +
>> +/* Reference Oscillator operations */
>> +static int roclk_endisable(struct clk_hw *hw, int enable)
>> +{
>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> +
>> + if (enable)
>> + clk_writel(REFO_ON | REFO_OE, PIC32_SET(refo->regs));
>> + else
>> + clk_writel(REFO_ON | REFO_OE, PIC32_CLR(refo->regs));
>> + return 0;
>> +}
>> +
>> +static int roclk_is_enabled(struct clk_hw *hw)
>> +{
>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> +
>> + return clk_readl(refo->regs) & REFO_ON;
>> +}
>> +
>> +static int roclk_enable(struct clk_hw *hw)
>> +{
>> + return roclk_endisable(hw, 1);
>> +}
>> +
>> +static void roclk_disable(struct clk_hw *hw)
>> +{
>> + roclk_endisable(hw, 0);
>> + cpu_relax();
> What is this for? Usually cpu_relax() only goes in loops. If it's
> still needed, we need a comment above it.

Good catch. This is added to insert a nop cycle. As per datasheet CPU should not access
the clock register at the immediate next cycle of the clock-gating instruction.
Will remove it, as we are not really accessing the clock register just after clock gate.

>> +}
>> +
>> +static void roclk_init(struct clk_hw *hw)
>> +{
>> + roclk_disable(hw);
>> +}
>> +
>> +static u8 roclk_get_parent(struct clk_hw *hw)
>> +{
>> + u8 i = 0;
> Use int for iterator types. u8 is for when you really care about
> the size.

ack.

>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> + unsigned long v;
>> +
>> + v = clk_readl(refo->regs);
>> + v = (v >> REFO_SEL_SHIFT) & REFO_SEL_MASK;
>> +
>> + if (!refo->parent_idx)
>> + goto done;
>> +
>> + for (i = 0; i < clk_hw_get_num_parents(hw); i++)
>> + if (refo->parent_idx[i] == v)
>> + return (u8)i;
>> +done:
>> + return (u8)v;
>> +}
>> +
>> +static int roclk_set_parent(struct clk_hw *hw, u8 index)
>> +{
>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> + unsigned long v, flags;
>> + u8 new_idx, cur_idx, was_disabled = 1;
>> +
>> + new_idx = index;
>> + if (refo->parent_idx && (index < clk_hw_get_num_parents(hw)))
>> + new_idx = refo->parent_idx[index];
>> +
>> + /* sanity */
>> + v = clk_readl(refo->regs);
>> + cur_idx = v & REFO_SEL_MASK;
>> +
>> + if (unlikely(cur_idx == new_idx))
>> + return 0;
>> +
>> + /*
>> + * Note: clk-src switching is allowed only when module is not ACTIVE.
>> + * Module gets ACTIVE when enabled. So it meant set_parent() needs
>> + * clk-gating across the call.
>> + */
>> + if (roclk_is_enabled(hw)) {
>> + pr_warn("%s needs gated clock. Forcing.\n", __func__);
>> + roclk_disable(hw);
>> + was_disabled = 0;
>> + }
>> +
>> + /* wait until ACTIVE bit is zero */
>> + for (;;) {
> Can we put some timeout on this? I'd rather not have the hardware
> flip a bit minutes, hours, days later and we're sitting here
> waiting.

ack. Replace with iopoll timeout variant.

>> + v = clk_readl(refo->regs);
>> + if ((v & REFO_ACTIVE) == 0)
>> + break;
>> + }
>> +
>> + __clk_lock(flags);
>> +
>> + /* sysunlock */
>> + pic32_devcon_sysunlock();
>> +
>> + /* Calculate REFOCON register value */
>> + v = clk_readl(refo->regs);
>> + v &= ~(REFO_SEL_MASK << REFO_SEL_SHIFT);
>> + v |= (new_idx << REFO_SEL_SHIFT);
>> +
>> + /* Apply */
>> + clk_writel(v, refo->regs);
>> +
>> + /* syslock */
>> + pic32_devcon_syslock();
>> +
>> + /* enable module */
>> + clk_writel(REFO_ON | REFO_OE, PIC32_SET(refo->regs));
>> +
>> + __clk_unlock(flags);
>> +
>> + /* keep it disabled, if it was */
>> + if (was_disabled)
>> + clk_writel(REFO_ON, PIC32_CLR(refo->regs));
>> +
>> + return 0;
>> +}
>> +
>> +static unsigned long roclk_calc_rate(unsigned long parent_rate,
>> + u16 rodiv, u16 rotrim)
>> +{
>> + u64 rate64;
>> + u32 N;
>> +
>> + N = rodiv;
>> + /* fout = fin / [2 * {N + (M / 512)}]
>> + * = fin * 512 / [1024 * N + 2 * M]
>> + * = fin * 256 / (512 * N + M)
>> + * = (fin << 8) / ((N << 9) + M)
>> + */
>> + if (rotrim) {
>> + N = (N << 9) + rotrim;
>> + rate64 = parent_rate;
>> + rate64 <<= 8;
>> + do_div(rate64, N);
>> + } else {
>> + rate64 = parent_rate / (N << 1);
>> + }
>> + return (unsigned long)rate64;
>> +}
>> +
>> +static void roclk_calc_div_trim(unsigned long rate,
>> + unsigned long parent_rate,
>> + u16 *rodiv_p, u16 *rotrim_p)
>> +{
>> + u16 div, rotrim, rodiv;
> Scared of overflow. Why aren't these just int or long?

ack. Will update.

>> + u64 frac;
>> +
>> + /* Find integer approximation of floating-point arithmatic.
> s/arithmatic/arithmetic/ ?

ack. :)

>> + * fout = fin / [2 * {rodiv + (rotrim / 512)}] ... (1)
>> + * i.e. fout = fin / 2 * DIV
>> + * whereas DIV = rodiv + (rotrim / 512)
>> + *
>> + * Since kernel does not perform floating-point arithmatic so
>> + * (rotrim/512) will be zero. And DIV & rodiv will result same.
>> + *
>> + * ie. fout = (fin * 256) / [(512 * rodiv) + rotrim] ... from (1)
>> + * ie. rotrim = ((fin * 256) / fout) - (512 * DIV)
>> + */
>> + if (parent_rate <= rate) {
>> + div = 0;
>> + frac = 0;
>> + rodiv = 0;
>> + rotrim = 0;
>> + } else {
>> + div = parent_rate / (rate << 1);
>> + frac = parent_rate;
>> + frac <<= 8;
>> + do_div(frac, rate);
>> + frac -= (u64)(div << 9);
>> +
>> + rodiv = (div > REFO_DIV_MASK) ? REFO_DIV_MASK : div;
>> + rotrim = (frac >= REFO_TRIM_MAX) ? REFO_TRIM_MAX : (u16)frac;
> Useless cast.

ack. Will drop.

>> + }
>> +
>> + if (rodiv_p)
>> + *rodiv_p = rodiv;
>> +
>> + if (rotrim_p)
>> + *rotrim_p = rotrim;
>> +}
>> +
>> +static unsigned long roclk_recalc_rate(struct clk_hw *hw,
>> + unsigned long parent_rate)
>> +{
>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> + unsigned long v;
>> + u16 rodiv, rotrim;
>> +
>> + /* get rodiv */
>> + v = clk_readl(refo->regs);
>> + rodiv = (v >> REFO_DIV_SHIFT) & REFO_DIV_MASK;
>> +
>> + /* get trim */
>> + v = clk_readl(refo->regs + REFO_TRIM_REG);
>> + rotrim = (v >> REFO_TRIM_SHIFT) & REFO_TRIM_MASK;
>> +
>> + v = roclk_calc_rate(parent_rate, rodiv, rotrim);
>> + return v;
> Just return roclk_calc_rate(...)

ack.

>> +}
>> +
>> +static long roclk_round_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long *parent_rate)
>> +{
>> + u16 rotrim, rodiv;
>> +
>> + /* calculate dividers for new rate */
>> + roclk_calc_div_trim(rate, *parent_rate, &rodiv, &rotrim);
>> +
>> + /* caclulate new rate (rounding) based on new rodiv & rotrim */
>> + return roclk_calc_rate(*parent_rate, rodiv, rotrim);
>> +}
>> +
>> +static int roclk_determine_rate(struct clk_hw *hw,
>> + struct clk_rate_request *req)
>> +{
>> + struct clk_hw *parent_clk, *best_parent_clk = NULL;
>> + unsigned int i, delta, best_delta = -1;
>> + unsigned long parent_rate, best_parent_rate = 0;
>> + unsigned long best = 0, nearest_rate;
>> +
>> + /* find a parent which can generate nearest clkrate >= rate */
>> + for (i = 0; i < clk_hw_get_num_parents(hw); i++) {
>> + /* get parent */
>> + parent_clk = clk_hw_get_parent_by_index(hw, i);
>> + if (!parent_clk)
>> + continue;
>> +
>> + /* skip if parent runs slower than target rate */
>> + parent_rate = clk_hw_get_rate(parent_clk);
>> + if (req->rate > parent_rate)
>> + continue;
>> +
>> + nearest_rate = roclk_round_rate(hw, req->rate, &parent_rate);
>> + delta = abs(nearest_rate - req->rate);
>> + if ((nearest_rate >= req->rate) && (delta < best_delta)) {
>> + best_parent_clk = parent_clk;
>> + best_parent_rate = parent_rate;
>> + best = nearest_rate;
>> + best_delta = delta;
>> +
>> + if (delta == 0)
>> + break;
>> + }
>> + }
> Why can't we use the generic mux determine rate code here?

Generic __clk_mux_determine_rate() and sister works with clocks which has
only muxing property and no divider, no multiplier. It selects best parents
which can generate frequency nearest to target rate.
But in PIC32 refo clock has (integer & floating-point) divider and input
selector which are capable of generating wide range of output frequencies
from the same parent clock. This custom .determine_rate() callback searches all
the parents to generate best output clock by applying different dividers.

>> +
>> + /* if no match found, retain old rate */
>> + if (!best_parent_clk) {
>> + pr_err("%s:%s, no parent found for rate %lu.\n",
>> + __func__, clk_hw_get_name(hw), req->rate);
>> + best_parent_clk = clk_hw_get_parent(hw);
>> + best_parent_rate = clk_hw_get_rate(best_parent_clk);
>> + best = clk_hw_get_rate(hw);
>> + }
>> +
>> + pr_debug("%s,rate %lu /best_parent(%s, %lu) /best %lu /delta %d\n",
>> + clk_hw_get_name(hw), req->rate,
>> + clk_hw_get_name(best_parent_clk), best_parent_rate,
>> + best, best_delta);
>> +
>> + if (req->best_parent_rate)
>> + req->best_parent_rate = best_parent_rate;
>> +
>> + if (req->best_parent_hw)
>> + req->best_parent_hw = best_parent_clk;
>> +
>> + return best;
>> +}
>> +
>> +static int roclk_set_rate_and_parent(struct clk_hw *hw,
>> + unsigned long rate,
>> + unsigned long parent_rate,
>> + u8 index)
>> +{
>> + struct pic32_refosc *refo = clkhw_to_refosc(hw);
>> + u16 trim, rodiv, parent_id, was_disabled = 1;
>> + unsigned long flags, v;
>> +
>> + if (unlikely(clk_hw_get_rate(hw) == rate))
>> + return 0;
>> +
>> + /* calculate new rodiv & rotrim for new rate */
>> + roclk_calc_div_trim(rate, parent_rate, &rodiv, &trim);
>> +
>> + pr_debug("parent_rate = %lu, rate = %lu, div = %d, trim = %d\n",
>> + parent_rate, rate, rodiv, trim);
>> +
>> + /* Note: rosel can only be programmed when module is INACTIVE.
>> + * i.e gating is required across set_parent.
>> + * So disable clk, if required.
>> + */
>> + if (roclk_is_enabled(hw)) {
>> + pr_err("%s: needs gating. Forcing.\n", __func__);
> We have a flag CLK_SET_RATE_GATE for this.

Ack. Will use CLK_SET_RATE_GATE, CLK_SET_PARENT_GATE accordingly.

>> + roclk_disable(hw);
>> + was_disabled = 0;
>> + }
>> +
>> + /* check current source */
>> + if (refo->parent_idx)
>> + index = refo->parent_idx[index];
>> +
>> + parent_id = roclk_get_parent(hw);
>> + if (parent_id == index)
>> + goto clk_rosel_ready;
>> +
>> + /* wait till source change is active */
>> + for (;;) {
>> + v = clk_readl(refo->regs);
>> + if ((v & (REFO_DIVSW_EN | REFO_ACTIVE)) == 0)
>> + break;
>> + }
>> +
>> +clk_rosel_ready:
>> + /* spinlock */
>> + __clk_lock(flags);
> If we spelled out spin_lock_irqsave() here instead we wouldn't
> need the comment. Please do that and get rid of any spinlock
> comments.

Ack, Will remove the wrapper.

>> + v = clk_readl(refo->regs);
>> +
>> + /* sysunlock */
>> + pic32_devcon_sysunlock();
>> +
>> + /* apply parent, if required */
>> + if (parent_id != index) {
>> + v &= ~(REFO_SEL_MASK << REFO_SEL_SHIFT);
>> + v |= (index << REFO_SEL_SHIFT);
>> + }
>> +
>> + /* apply RODIV */
>> + v &= ~(REFO_DIV_MASK << REFO_DIV_SHIFT);
>> + v |= (rodiv << REFO_DIV_SHIFT);
>> + clk_writel(v, refo->regs);
>> +
>> + /* apply ROTRIM */
>> + v = clk_readl(refo->regs + REFO_TRIM_REG);
>> + v &= ~(REFO_TRIM_MASK << REFO_TRIM_SHIFT);
>> + v |= (trim << REFO_TRIM_SHIFT);
>> + clk_writel(v, refo->regs + REFO_TRIM_REG);
>> +
>> + /* enable refo module */
>> + clk_writel(REFO_ON | REFO_OE, PIC32_SET(refo->regs));
>> +
>> + /* activate divider switching */
>> + clk_writel(REFO_DIVSW_EN, PIC32_SET(refo->regs));
>> +
>> + /* syslock */
> Useless comment.

ack.

>> + pic32_devcon_syslock();
>> +
>> + /* wait till divswen is in-progress */
>> + for (;;) {
>> + v = clk_readl(refo->regs);
>> + if ((v & REFO_DIVSW_EN) == 0)
>> + break;
>> + }
>> +
>> + __clk_unlock(flags);
>> +
>> + /* keep it disabled if it was */
>> + if (was_disabled)
>> + clk_writel(REFO_ON, PIC32_CLR(refo->regs));
>> +
>> + return 0;
>> +}
>> +
>> +static int roclk_set_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long parent_rate)
>> +{
>> + u8 index = roclk_get_parent(hw);
>> +
>> + return roclk_set_rate_and_parent(hw, rate, parent_rate, index);
>> +}
>> +
>> +static struct debugfs_reg32 roclk_regs_debug[] = {
>> + { .name = "REFOxCON", .offset = 0,},
>> + { .name = "REFOxTRIM", .offset = REFO_TRIM_REG,},
>> +};
>> +
>> +static int roclk_debug_init(struct clk_hw *hw, struct dentry *dentry)
>> +{
>> + struct pic32_refosc *ro = clkhw_to_refosc(hw);
>> + struct dentry *file;
>> +
>> + ro->regset.base = ro->regs;
>> + ro->regset.regs = roclk_regs_debug;
>> + ro->regset.nregs = ARRAY_SIZE(roclk_regs_debug);
>> +
>> + file = debugfs_create_regset32("regdump", S_IRUGO, dentry, &ro->regset);
>> +
>> + return IS_ERR(file) ? PTR_ERR(file) : 0;
> Wrong check for error.

ack.

>> +}
>> +
>> +static inline u8 spll_odiv_to_divider(u8 odiv)
>> +{
>> + if (odiv <= PLL_ODIV_MIN)
>> + odiv = PLL_ODIV_MIN;
>> + else if (odiv >= PLL_ODIV_MAX)
>> + odiv = PLL_ODIV_MAX;
> clamp(odiv, PLL_ODIV_MIN, PLL_ODIV_MAX)

ack.

>> +
>> + return 1 << odiv;
>> +}
>> +
>> +static unsigned long spll_calc_mult_div(struct pic32_spll *pll,
>> + unsigned long rate,
>> + unsigned long parent_rate,
>> + u8 *mult_p, u8 *odiv_p)
>> +{
>> + u8 mul, div, best_mul = 1, best_div = 1;
>> + unsigned long new_rate, best_rate = rate;
>> + unsigned int best_delta = -1, delta, match_found = 0;
>> + u64 rate64;
>> +
>> + parent_rate /= pll->idiv;
>> +
>> + for (mul = 1; mul <= PLL_MULT_MAX; mul++) {
>> + for (div = PLL_ODIV_MIN; div <= PLL_ODIV_MAX; div++) {
>> + rate64 = parent_rate;
>> + rate64 *= mul;
>> + do_div(rate64, 1 << div);
>> + new_rate = (u32)rate64;
>> + delta = abs(rate - new_rate);
>> + if ((new_rate >= rate) && (delta < best_delta)) {
>> + best_delta = delta;
>> + best_rate = new_rate;
>> + best_mul = mul;
>> + best_div = div;
>> + match_found = 1;
>> + }
>> + }
>> + }
>> +
>> + if (!match_found) {
>> + pr_warn("spll: no match found\n");
>> + return 0;
>> + }
>> +
>> + pr_debug("rate %lu, par_rate %lu/mult %u, div %u, best_rate %lu\n",
>> + rate, parent_rate, best_mul, best_div, best_rate);
>> +
>> + if (mult_p)
>> + *mult_p = best_mul - 1;
>> +
>> + if (odiv_p)
>> + *odiv_p = best_div;
>> +
>> + return best_rate;
>> +}
>> +
>> +static unsigned long spll_clk_recalc_rate(struct clk_hw *hw,
>> + unsigned long parent_rate)
>> +{
>> + struct pic32_spll *pll = clkhw_to_spll(hw);
>> + unsigned long pll_in_rate, v;
>> + u8 mult, odiv, div;
>> + u64 rate64;
>> +
>> + v = clk_readl(pll->regs);
>> + odiv = ((v >> PLL_ODIV_SHIFT) & PLL_ODIV_MASK);
>> + mult = ((v >> PLL_MULT_SHIFT) & PLL_MULT_MASK) + 1;
> mult can't overflow here?

ack, As per datasheet 'mult' is 7-bit wide. So will not overflow. Anyway will promote it to u32.

>> + div = spll_odiv_to_divider(odiv);
>> +
>> + /* pll_in = parent_rate / idiv
>> + * pll_out = pll_in * mult / div;
>> + */
>> + pll_in_rate = parent_rate / pll->idiv;
>> + rate64 = pll_in_rate;
>> + rate64 *= mult;
>> + do_div(rate64, div);
>> +
>> + return (unsigned long)rate64;
>> +}
>> +
>> +static long spll_clk_round_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long *parent_rate)
>> +{
>> + struct pic32_spll *pll = clkhw_to_spll(hw);
>> +
>> + return spll_calc_mult_div(pll, rate, *parent_rate, NULL, NULL);
>> +}
>> +
>> +static int spll_clk_set_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long parent_rate)
>> +{
>> + struct pic32_spll *pll = clkhw_to_spll(hw);
>> + u8 mult, odiv;
>> + unsigned long ret, v, loop = 1000;
>> + struct clk *sclk_parent;
>> + unsigned long flags;
>> +
>> + ret = spll_calc_mult_div(pll, rate, parent_rate, &mult, &odiv);
>> + if (!ret || (ret == rate))
> Drop useless parenthesis.

ack.

>> + return 0;
>> +
>> + /* To change frequency
>> + * - (a) check whether this clk is active parent of SYSCLK.
>> + * - (b) apply new mult & odiv.
>> + * - (c) switch back to PLL
>> + * - (d) wait until PLL settles down / locked.
> a,b,c,d is usually used for a choice. Please use 1,2,3,4 instead.

ack.

>> + */
>> +
>> + /* To check whether rate change is allowed we will have to ensure
>> + * spll_clk is not active parent of sys_clk.
>> + */
>> + if (WARN_ON(IS_ERR_OR_NULL(pic32_sys_clk)))
> Does this ever occur? We control when pic32_sys_clk is assigned
> so it seems like all we could have is broken code.

This is not required at all. Will remove.

>> + return -EPERM;
>> +
>> + /* get sysclk parent */
>> + sclk_parent = clk_get_parent(pic32_sys_clk);
> Ah here's the clk API usage. Maybe we need some clk_hw API that
> tells us if a clk_hw is a child of this clk_hw? Something like:
>
> bool clk_hw_is_child(struct clk_hw *hw, struct clk_hw *child)
>
> This also highlights a problem we have with CLK_SET_RATE_GATE,
> where that flag isn't considered when the clock is changing rate
> due to something upstream. We may want to add some new flag, or
> just change the behavior so that CLK_SET_RATE_GATE is always
> respected.

In PIC32 one of the possible "immediate" parent of SCLK is SPLL, so I hope comparing
clk_hw_get_parent(sclk_hw) with spll_hw is sufficient to find parent-child relation.

In this function 'spll_clk_set_rate()' we want to ensure correct pre-condition for
SPLL rate change is followed, in-case of condition-check failure we will return with error,
else will apply the new rate.
Note we will not force set_parent() or re-parent from this function because of internal
complexity in handling flow.
So clock user can follow this sequence to perform SPLL rate change:
clk_set_parent(sclk, frcdiv_clk);
clk_set_rate(spll, new-rate);
clk_set_parent(sclk, spll);

Will update comment accordingly.

>> +
>> + /* does sys_clk using spll_clk as parent ? */
> Is sys_clk using spll_clk as its parent?
>
>> + if (unlikely(__clk_get_hw(sclk_parent) == hw)) {
>> + pr_err("spll: set_rate() is not allowed when spll is parent of sys_clk.");
>> + pr_err("First reparent sys_clk to frcdiv-clk and then try.\n");
> This will look like "clk.First reparent". Is that intentional?
> And is the user looking at the kernel log even going to know what
> that means?

Will remove.

>> + return -EPERM;
>> + }
>> +
>> + /* lock */
>> + __clk_lock(flags);
>> +
>> + /* apply new multiplier & divisor (read-modify-write) */
>> + v = clk_readl(pll->regs);
>> + v &= ~(PLL_MULT_MASK << PLL_MULT_SHIFT);
>> + v &= ~(PLL_ODIV_MASK << PLL_ODIV_SHIFT);
>> + v |= (mult << PLL_MULT_SHIFT) | (odiv << PLL_ODIV_SHIFT);
>> +
>> + /* sysunlock before writing to SPLLCON register */
>> + pic32_devcon_sysunlock();
>> +
>> + clk_writel(v, pll->regs);
>> + cpu_relax();
>> +
>> + /* insert few nops (5-stage) to ensure CPU does not hang */
>> + cpu_nop5();
>> + cpu_nop5();
>> +
>> + /* syslock*/
>> + pic32_devcon_syslock();
>> +
>> + /* Wait until PLL is locked (maximum 100 usecs). */
>> + for (;;) {
>> + v = clk_readl(pll->status_reg);
>> + if (v & pll->pll_locked)
>> + break;
>> +
>> + if (--loop == 0)
>> + break;
>> +
>> + ndelay(100);
>> + }
>> +
>> + /* lock */
>> + __clk_unlock(flags);
>> +
>> + return 0;
>> +}
>> +
>> +static struct debugfs_reg32 sclk_regs_debug[] = {
>> + { .name = "OSCCON", .offset = 0,},
>> + { .name = "OSCTUN", .offset = 0x10,},
>> + { .name = "SPLLCON", .offset = 0x20,},
>> +};
>> +
>> +static int sclk_debug_init(struct clk_hw *hw, struct dentry *dir)
>> +{
>> + struct pic32_sclk *sclk = clkhw_to_sys_clk(hw);
>> + struct dentry *file;
>> +
>> + sclk->regset.base = sclk->regs;
>> + sclk->regset.regs = sclk_regs_debug;
>> + sclk->regset.nregs = ARRAY_SIZE(sclk_regs_debug);
>> +
>> + file = debugfs_create_regset32("regdump", S_IRUGO, dir, &sclk->regset);
>> + if (IS_ERR(file))
> Debugfs APIs check for NULL or non-NULL to indicate error or
> success.

ack. Will drop as mentioned above.

>> + return PTR_ERR(file);
>> +
>> + return 0;
> Can be simplified to return PTR_ERR_OR_ZERO().
>
> Also, have you considered using regmap-mmio?
>
>> +}
>> +
>> +static long sclk_round_rate(struct clk_hw *hw, unsigned long rate,
>> + unsigned long *parent_rate)
>> +{
>> + return calc_best_divided_rate(rate, *parent_rate, SLEW_SYSDIV, 1);
>> +}
>> +
>> +static unsigned long sclk_get_rate(struct clk_hw *hw, unsigned long parent_rate)
>> +{
>> + u32 v, div;
>> + struct pic32_sclk *sysclk = clkhw_to_sys_clk(hw);
>> +
>> + v = clk_readl(sysclk->slwreg);
>> + div = (v >> SLEW_SYSDIV_SHIFT) & SLEW_SYSDIV;
>> + div += 1; /* sys-div to divider */
>> +
>> + return parent_rate / div;
>> +}
>> +
>> +static int sclk_set_rate(struct clk_hw *hw,
>> + unsigned long rate, unsigned long parent_rate)
>> +{
>> + u32 v, div;
>> + unsigned long flags;
>> + struct pic32_sclk *sysclk = clkhw_to_sys_clk(hw);
>> + ktime_t timeout;
>> +
>> + div = parent_rate / rate;
>> +
>> + __clk_lock(flags);
>> +
>> + /* sysunlock*/
>> + pic32_devcon_sysunlock();
>> +
>> + /* apply new div */
>> + v = clk_readl(sysclk->slwreg);
>> + v &= ~(SLEW_SYSDIV << SLEW_SYSDIV_SHIFT);
>> + v |= ((div - 1) << SLEW_SYSDIV_SHIFT);
>> + clk_writel(v, sysclk->slwreg);
>> +
>> + /* syslock*/
>> + pic32_devcon_syslock();
>> +
>> + /* wait until BUSY is cleared */
>> + timeout = ktime_add_ns(ktime_get(), LOCK_TIMEOUT_NS);
>> + for (;;) {
>> + v = clk_readl(sysclk->slwreg);
>> + if (!(v & SLEW_BUSY))
>> + break;
>> +
>> + if (ktime_after(ktime_get(), timeout)) {
>> + pr_err("%s: busy while timeout\n",
>> + clk_hw_get_name(hw));
>> + break;
>> + }
>> + }
>> + __clk_unlock(flags);
>> +
>> + return 0;
>> +}
>> +
>> +static u8 sclk_get_parent(struct clk_hw *hw)
>> +{
>> + u8 idx, i;
>> + u32 v;
>> + struct pic32_sclk *sysclk = clkhw_to_sys_clk(hw);
>> +
>> + v = clk_readl(sysclk->regs);
>> + idx = (v >> OSC_CUR_SHIFT) & OSC_CUR_MASK;
>> +
>> + if (!sysclk->parent_idx)
>> + goto done;
> Just return idx then.

ack.

>> +
>> + for (i = 0; i < clk_hw_get_num_parents(hw); i++) {
>> + if (sysclk->parent_idx[i] == idx) {
>> + idx = i;
>> + break;
>> + }
>> + }
>> +
>> +done:
>> + return idx;
>> +}
>> +
>> +static int sclk_set_parent(struct clk_hw *hw, u8 index)
>> +{
>> + u32 v;
>> + unsigned long flags, parent_rate;
>> + u8 nosc, cosc;
>> + struct pic32_sclk *sysclk = clkhw_to_sys_clk(hw);
>> +
>> + /* find new_osc */
>> + nosc = sysclk->parent_idx ? sysclk->parent_idx[index] : index;
>> +
>> + /* check cur_osc is not same as new_osc */
>> + v = clk_readl(sysclk->regs);
>> + cosc = (v >> OSC_CUR_SHIFT) & OSC_CUR_MASK;
>> + if (unlikely(cosc == nosc))
>> + return 0;
>> +
>> + parent_rate = clk_hw_get_rate(clk_hw_get_parent_by_index(hw, index));
>> +
>> + /* spin lock */
>> + __clk_lock(flags);
>> +
>> + /* sysunlock*/
>> + pic32_devcon_sysunlock();
>> +
>> + /* set new parent */
>> + v = clk_readl(sysclk->regs);
>> + v &= ~(OSC_NEW_MASK << OSC_NEW_SHIFT);
>> + v |= (nosc << OSC_NEW_SHIFT);
> Drop useless parenthesis.

ack.

>> + clk_writel(v, sysclk->regs);
>> +
>> + /* initate switch */
>> + clk_writel(OSC_SWEN, PIC32_SET(sysclk->regs));
>> + cpu_relax();
>> +
>> + /* some nop to flush pipeline(cpu-clk is in-flux) */
>> + cpu_nop5();
>> +
>> + /* syslock */
>> + pic32_devcon_syslock();
>> +
>> + /* wait for SWEN bit to clear */
>> + for (;;) {
>> + v = clk_readl(sysclk->regs);
>> + if (!(v & OSC_SWEN))
>> + break;
>> + }
>> +
>> + /* spin unlock */
>> + __clk_unlock(flags);
>> +
>> + /* SYSCLK switch logic performs sanity and maintains state machine for
>> + * clock-switching. So h/w might reject clk-switch request if required
>> + * conditions (like clksrc not present or unstable) aren't met.
>> + * So confirm before claiming success.
>> + */
>> + cosc = (v >> OSC_CUR_SHIFT) & OSC_CUR_MASK;
>> + if (unlikely(cosc != nosc)) {
>> + pr_err("%s: err COSC %d and NOSC %d\n",
>> + clk_hw_get_name(hw), cosc, nosc);
>> + return -EBUSY;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int sosc_clk_enable(struct clk_hw *hw)
>> +{
>> + int loop;
>> + unsigned long flags;
>> + struct pic32_sosc *sosc = clkhw_to_sosc(hw);
>> +
>> + local_irq_save(flags);
> This is already called with irqs saved and disabled, so what's
> the point of doing that again here?

ack. Will remove.

>> +
>> + /* enable SOSC */
>> + pic32_devcon_sysunlock();
>> + clk_writel(sosc->bitmask, PIC32_SET(sosc->regs));
>> + pic32_devcon_syslock();
>> +
>> + /* Wait till warm-up period expires and ready-status is updated */
>> + for (loop = 1024; loop; --loop) {
>> + cpu_relax();
>> + if (clk_readl(sosc->status_reg) & sosc->status_bitmask)
>> + break;
>> + }
>> +
>> + local_irq_restore(flags);
>> +
>> + if (!loop) {
>> + pr_err("%s: possibly clk is not present or ready for ops\n",
>> + clk_hw_get_name(hw));
>> + return -EBUSY;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static void sosc_clk_disable(struct clk_hw *hw)
>> +{
>> + struct pic32_sosc *sosc = clkhw_to_sosc(hw);
>> + unsigned long flags;
>> +
>> + local_irq_save(flags);
> Same story.

Will remove.

>> +
>> + pic32_devcon_sysunlock();
>> + clk_writel(sosc->bitmask, PIC32_CLR(sosc->regs));
>> + pic32_devcon_syslock();
>> +
>> + local_irq_restore(flags);
>> +}
>> +
>> +static int sosc_clk_is_enabled(struct clk_hw *hw)
>> +{
>> + struct pic32_sosc *sosc = clkhw_to_sosc(hw);
>> + u32 enable, status;
>> +
>> + /* check enable & ready-status */
>> + enable = clk_readl(sosc->regs) & sosc->bitmask;
>> + status = clk_readl(sosc->status_reg) & sosc->status_bitmask;
>> +
>> + return enable && status;
>> +}
>> +
>> +static unsigned long sosc_clk_calc_rate(struct clk_hw *hw,
>> + unsigned long parent_rate)
>> +{
>> + return clkhw_to_sosc(hw)->fixed_rate;
>> +}
>> +
>> +static struct clk_ops pbclk_ops = {
> Can this be const?

ack.

>> + .enable = pbclk_enable,
>> + .disable = pbclk_disable,
>> + .is_enabled = pbclk_is_enabled,
>> + .recalc_rate = pbclk_recalc_rate,
>> + .round_rate = pbclk_round_rate,
>> + .set_rate = pbclk_set_rate,
>> + .debug_init = pbclk_debug_init,
>> +};
>> +
>> +/* sysclk is a mux with post-divider.
>> + * get/set_parent & get/set_rate are required operation.
>> + */
>> +static struct clk_ops sclk_postdiv_ops = {
>> + .get_parent = sclk_get_parent,
>> + .set_parent = sclk_set_parent,
>> + .determine_rate = __clk_mux_determine_rate,
>> + .round_rate = sclk_round_rate,
>> + .set_rate = sclk_set_rate,
>> + .recalc_rate = sclk_get_rate,
>> + .debug_init = sclk_debug_init,
>> +};
>> +
>> +static struct clk_ops spll_clk_ops = {
>> + .recalc_rate = spll_clk_recalc_rate,
>> + .round_rate = spll_clk_round_rate,
>> + .set_rate = spll_clk_set_rate,
>> +};
>> +
>> +static struct clk_ops roclk_ops = {
>> + .enable = roclk_enable,
>> + .disable = roclk_disable,
>> + .is_enabled = roclk_is_enabled,
>> + .get_parent = roclk_get_parent,
>> + .set_parent = roclk_set_parent,
>> + .determine_rate = roclk_determine_rate,
>> + .recalc_rate = roclk_recalc_rate,
>> + .round_rate = roclk_round_rate,
>> + .set_rate_and_parent = roclk_set_rate_and_parent,
>> + .set_rate = roclk_set_rate,
>> + .init = roclk_init,
>> + .debug_init = roclk_debug_init,
>> +};
>> +
>> +static struct clk_ops sosc_ops = {
>> + .enable = sosc_clk_enable,
>> + .disable = sosc_clk_disable,
>> + .is_enabled = sosc_clk_is_enabled,
>> + .recalc_rate = sosc_clk_calc_rate,
>> +};
> Same const question for all these ops.

ack.

>> +
>> +#define init_clk_data(__initdata, __clk, __parents, \
>> + __nr_parents, __flags, __ops) \
>> + __initdata.name = (__clk); \
>> + __initdata.ops = (__ops); \
>> + __initdata.flags = (__flags); \
>> + __initdata.parent_names = (__parents); \
>> + __initdata.num_parents = (__nr_parents)
>> +
>> +static struct clk *periph_clk_register(const char *name,
>> + const char **parent_name,
>> + void __iomem *regs, u32 flags)
>> +{
>> + struct clk *clk;
>> + struct pic32_pbclk *pbclk;
>> + struct clk_init_data init;
>> +
>> + init_clk_data(init, name, parent_name, 1,
>> + flags | CLK_IS_BASIC, &pbclk_ops);
> Don't use CLK_IS_BASIC.

ack. Will drop here and in other places too.

>> +
>> + pbclk = kzalloc(sizeof(*pbclk), GFP_KERNEL);
>> + if (!pbclk)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + /* init */
>> + pbclk->regs = regs;
>> + pbclk->flags = flags;
>> + pbclk->hw.init = &init;
>> +
>> + clk = clk_register(NULL, &pbclk->hw);
>> + if (IS_ERR(clk))
>> + kfree(pbclk);
>> +
>> + return clk;
>> +}
>> +
>> +static struct clk *sys_mux_clk_register(const char *name,
>> + const char **parents,
>> + const int num_parents,
>> + void __iomem *regs,
>> + void __iomem *slew_reg,
>> + u32 *parent_idx,
>> + const struct clk_ops *clkop)
>> +{
>> + struct clk *clk;
>> + struct pic32_sclk *sysclk;
>> + struct clk_init_data init;
>> +
>> + init_clk_data(init, name, parents, num_parents,
>> + CLK_IS_BASIC, clkop);
>> +
>> + sysclk = kzalloc(sizeof(*sysclk), GFP_KERNEL);
>> + if (!sysclk)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + /* init sysclk data */
>> + sysclk->hw.init = &init;
>> + sysclk->regs = regs;
>> + sysclk->slwreg = slew_reg;
>> + sysclk->parent_idx = parent_idx;
>> +
>> + clk = clk_register(NULL, &sysclk->hw);
>> + if (IS_ERR(clk)) {
>> + kfree(sysclk);
>> + return clk;
>> + }
>> +
>> + /* Maintain reference to this clock;
>> + * This clock will be needed in spll-rate-change.
> Please do multi-line comments with a /* on its own line.

ack.

>> + */
>> + pic32_sys_clk = clk;
>> +
>> + return clk;
>> +}
>> +
>> +static struct clk *spll_clk_register(const char *name, const char *parents,
>> + void __iomem *regs,
>> + void __iomem *status_reg,
>> + u32 lock_bitmask)
>> +{
>> + u32 v;
>> + struct pic32_spll *pll;
>> + struct clk_init_data init;
>> + struct clk *clk;
>> +
>> + init_clk_data(init, name, &parents, 1, CLK_IS_BASIC, &spll_clk_ops);
>> +
>> + pll = kzalloc(sizeof(*pll), GFP_KERNEL);
>> + if (!pll)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + /* initialize configuration */
>> + pll->regs = regs;
>> + pll->status_reg = status_reg;
>> + pll->pll_locked = lock_bitmask;
>> + pll->hw.init = &init;
>> +
>> + /* read and cache pll_idiv; we will use it as constant.*/
>> + v = clk_readl(pll->regs);
>> + pll->idiv = ((v >> PLL_IDIV_SHIFT) & PLL_IDIV_MASK) + 1;
>> +
>> + clk = clk_register(NULL, &pll->hw);
>> + if (IS_ERR(clk))
>> + kfree(pll);
>> +
>> + return clk;
>> +}
>> +
>> +static struct clk *refo_clk_register(const char *name,
>> + const char **parents,
>> + u32 nr_parents,
>> + void __iomem *regs,
>> + u32 *parent_idx)
>> +{
>> + struct pic32_refosc *refo;
>> + struct clk_init_data init;
>> + struct clk *clk;
>> + int clk_flags = CLK_IS_BASIC;
>> +
>> + init_clk_data(init, name, parents, nr_parents, clk_flags, &roclk_ops);
>> +
>> + refo = kmalloc(sizeof(*refo), GFP_KERNEL);
>> + if (!refo)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + /* initialize configuration */
>> + refo->regs = regs;
>> + refo->hw.init = &init;
>> + refo->parent_idx = parent_idx;
>> +
>> + clk = clk_register(NULL, &refo->hw);
>> + if (IS_ERR(clk))
>> + kfree(refo);
>> +
>> + return clk;
>> +}
>> +
>> +static void __init of_sosc_clk_setup(struct device_node *np)
>> +{
>> + u32 rate, stsmask, bitmask;
>> + struct pic32_sosc *sosc;
>> + struct clk *clk;
>> + struct clk_init_data init;
>> + void __iomem *regs, *status_reg;
>> + const char *name = np->name;
>> +
>> + if (of_property_read_u32(np, "clock-frequency", &rate))
>> + return;
>> +
>> + of_property_read_string(np, "clock-output-names", &name);
>> +
>> + regs = of_iomap(np, 0);
>> + if (!regs)
>> + regs = pic32_clk_regbase;
>> +
>> + status_reg = of_iomap(np, 1);
>> + if (!status_reg)
>> + status_reg = regs;
>> +
>> + of_property_read_u32(np, "microchip,bit-mask", &bitmask);
>> +
>> + of_property_read_u32(np, "microchip,status-bit-mask", &stsmask);
>> +
>> + /* allocate fixed-rate clock */
> Uhh yes.
>
>> + sosc = kzalloc(sizeof(*sosc), GFP_KERNEL);
>> + if (!sosc)
>> + return;
>> +
>> + init_clk_data(init, name, NULL, 0,
>> + CLK_IS_BASIC | CLK_IS_ROOT, &sosc_ops);
>> +
>> + /* struct clk assignments */
>> + sosc->fixed_rate = rate;
>> + sosc->hw.init = &init;
>> + sosc->regs = regs;
>> + sosc->status_reg = status_reg;
>> + sosc->bitmask = bitmask;
>> + sosc->status_bitmask = stsmask;
>> +
>> + /* register the clock */
>> + clk = clk_register(NULL, &sosc->hw);
>> + if (IS_ERR(clk))
>> + kfree(sosc);
>> + else
>> + pic32_of_clk_register_clkdev(np, clk);
>> +}
>> +
>> +static void __init of_periph_clk_setup(struct device_node *np)
>> +{
>> + const char *parent_name;
>> + const char *name = np->name;
>> + struct clk *clk;
>> + u32 flags = 0;
>> + void __iomem *regs;
>> +
>> + regs = of_iomap(np, 0);
>> + if (!regs) {
>> + pr_err("%s: could not get reg property\n", name);
>> + return;
>> + }
>> +
>> + parent_name = of_clk_get_parent_name(np, 0);
>> + if (!parent_name) {
>> + pr_err("pbclk: %s must have a parent\n", name);
>> + goto err_map;
>> + }
>> +
>> + if (of_find_property(np, "microchip,ignore-unused", NULL)) {
> The bindings have really been acked? I think we've been actively
> prevented from putting linux concepts into DT, and ignore-unused
> is one of those.

Property 'microchip,ignore-unused' is to identify critical clocks (like CPU clock)
in the system which should never be gated for whatsoever be the reason. I'll drop
this now and use 'critical-clock' whenever available in upstream.

>> + flags |= CLK_IGNORE_UNUSED;
>> + pr_info("%s: ignore gating even if unused\n", name);
>> + }
>> +
>> + of_property_read_string(np, "clock-output-names", &name);
>> +
>> + /* register peripheral clock */
> This comment is worthless.

ack.

>> + clk = periph_clk_register(name, &parent_name, regs, flags);
>> + if (IS_ERR(clk)) {
>> + pr_err("%s: could not register clock\n", name);
>> + goto err_map;
>> + }
>> +
>> + pic32_of_clk_register_clkdev(np, clk);
>> +
>> + return;
>> +
>> +err_map:
>> + iounmap(regs);
>> +}
>> +
>> +static void __init of_refo_clk_setup(struct device_node *np)
>> +{
>> + struct clk *clk;
>> + int ret, count;
>> + const char **parents;
>> + const char *clk_name = np->name;
>> + void __iomem *regs;
>> + u32 *parent_idx;
>> +
>> + /* get the input clock source count */
>> + count = of_clk_get_parent_count(np);
>> + if (count < 0) {
>> + pr_err("%s: get clock count error\n", np->name);
>> + return;
>> + }
>> +
>> + parents = kzalloc((sizeof(char *) * count), GFP_KERNEL);
> kcalloc.

ack.

>> + if (!parents)
>> + return;
>> +
>> + ret = pic32_of_clk_get_parent_indices(np, &parent_idx, count);
>> + if (ret)
>> + goto err_parent;
>> +
>> + of_clk_parent_fill(np, parents, count);
>> +
>> + /* get iobase */
>> + regs = of_iomap(np, 0);
>> + if (!regs) {
>> + pr_err("%s: could not get reg property\n", np->name);
>> + goto err_parent_idx;
>> + }
>> +
>> + of_property_read_string(np, "clock-output-names", &clk_name);
>> +
>> + clk = refo_clk_register(clk_name, parents, count, regs, parent_idx);
>> + if (IS_ERR(clk)) {
>> + pr_err("%s: could not register clock\n", clk_name);
>> + goto err_map;
>> + }
>> +
>> + pic32_of_clk_register_clkdev(np, clk);
>> +
>> + goto err_parent;
>> +
>> +err_map:
>> + iounmap(regs);
>> +err_parent_idx:
>> + kfree(parent_idx);
>> +err_parent:
>> + kfree(parents);
>> +}
>> +
>> +static void __init of_sys_mux_slew_setup(struct device_node *np)
>> +{
>> + struct clk *clk;
>> + int ret, count;
>> + const char *clk_name;
>> + const char **parents;
>> + u32 *parent_idx, slew, v;
>> + unsigned long flags;
>> + void __iomem *slew_reg;
>> +
>> + /* get the input clock source count */
>> + count = of_clk_get_parent_count(np);
>> + if (count < 0) {
>> + pr_err("%s: get clock count error\n", np->name);
>> + return;
>> + }
>> +
>> + parents = kzalloc((sizeof(char *) * count), GFP_KERNEL);
> kcalloc

ack.

>> + if (!parents)
>> + return;
>> +
>> + ret = pic32_of_clk_get_parent_indices(np, &parent_idx, count);
>> + if (ret)
>> + goto err_name;
>> +
>> + of_clk_parent_fill(np, parents, count);
>> +
>> + ret = of_property_read_string_index(np, "clock-output-names",
>> + 0, &clk_name);
>> + if (ret)
>> + clk_name = np->name;
> If at all possible, don't use clock-output-names to set the name
> of your clock. We're trying to figure out how to get rid of clk
> names being used by the clk framework to describe the hierarchy.

ack, will drop.

>> +
>> + /* get slew base */
>> + slew_reg = of_iomap(np, 0);
>> + if (!slew_reg) {
>> + pr_warn("%s: no slew register ?\n", clk_name);
> More like, couldn't map slew register, or just silence.

ack.

>> + goto err_name;
>> + }
>> +
>> + /* register mux clk */
>> + clk = sys_mux_clk_register(clk_name, parents, count, pic32_clk_regbase,
>> + slew_reg, parent_idx, &sclk_postdiv_ops);
>> + if (IS_ERR(clk)) {
>> + pr_err("%s: could not register clock\n", clk_name);
>> + goto err_parent_idx;
>> + }
>> +
>> + /* enable slew, if asked */
>> + if (!of_property_read_u32(np, "microchip,slew-step", &slew)) {
>> + __clk_lock(flags);
>> +
>> + v = clk_readl(slew_reg);
>> + /* Apply new slew-div and enable up/down slewing */
>> + v &= ~(SLEW_DIV << SLEW_DIV_SHIFT);
>> + v |= (slew << SLEW_DIV_SHIFT);
> Useless parenthesis.

ack. Will remove others as well.

>> + v |= SLEW_DOWNEN | SLEW_UPEN;
>> + clk_writel(v, slew_reg);
>> +
>> + __clk_unlock(flags);
>> + }
>> +
>> + /* register clkdev */
>> + pic32_of_clk_register_clkdev(np, clk);
>> +
>> + goto err_name;
>> +
>> +err_parent_idx:
>> + iounmap(slew_reg);
>> + kfree(parent_idx);
>> +err_name:
>> + kfree(parents);
>> +}
>> +
>> +static void __init of_sys_pll_setup(struct device_node *np)
>> +{
>> + int count;
>> + const char *clk_name = np->name;
>> + const char **parent_names;
>> + const char *plliclk_name = "spll_mux_clk";
>> + void __iomem *regs, *stat_reg;
>> + struct clk *clk, *mux_clk;
>> + u32 bitmask;
>> +
>> + /* get the input clock source count */
>> + count = of_clk_get_parent_count(np);
>> + if (count < 0) {
>> + pr_err("%s: get clock count error, %d\n", np->name, count);
>> + return;
>> + }
>> +
>> + parent_names = kzalloc((sizeof(char *) * count), GFP_KERNEL);
> kcalloc. I see a pattern.

ack.

>> + if (!parent_names)
>> + return;
>> +
>> + of_clk_parent_fill(np, parent_names, count);
>> +
>> + /* get output name */
>> + of_property_read_string(np, "clock-output-names", &clk_name);
>> +
>> + /* get iobase */
>> + regs = of_iomap(np, 0);
>> + if (!regs) {
>> + pr_err("%s: of_iomap failed\n", np->name);
>> + goto err_name;
>> + }
>> +
>> + /* get status reg & status bitmask */
>> + stat_reg = of_iomap(np, 1);
>> +
>> + of_property_read_u32(np, "microchip,status-bit-mask", &bitmask);
> Yuck. Bit masks in DT? This was approved?

Yes, this bitmask is from datasheet and accepted as border-line case.

>> + if (!stat_reg || !bitmask)
>> + pr_warn("%s: status_reg(or bit-mask) not found.\n", np->name);
>> +
>> + /* register plliclk mux */
>> + mux_clk = clk_register_mux(NULL, plliclk_name, parent_names,
>> + count, 0, regs,
>> + PLL_ICLK_SHIFT, 1, 0, &lock);
>> + if (IS_ERR(mux_clk)) {
>> + pr_err("splliclk_mux not registered\n");
>> + goto err_unmap;
>> + }
>> +
>> + /* register sys-pll clock */
>> + clk = spll_clk_register(clk_name, plliclk_name,
>> + regs, stat_reg, bitmask);
>> + if (IS_ERR(clk)) {
>> + pr_err("spll_clk not registered\n");
>> + goto err_mux;
>> + }
>> +
>> + pic32_of_clk_register_clkdev(np, clk);
>> + goto err_name;
>> +
>> +err_mux:
>> + clk_unregister(mux_clk);
>> +err_unmap:
>> + iounmap(regs);
>> +err_name:
>> + kfree(parent_names);
>> +}
>> +
>> +static void __init of_frcdiv_setup(struct device_node *np)
>> +{
>> + struct clk *clk;
>> + const char *clk_name = np->name;
>> + const char *parent_name;
>> +
>> + parent_name = of_clk_get_parent_name(np, 0);
>> + if (!parent_name) {
>> + pr_err("frcdiv: %s must have a parent\n", np->name);
>> + return;
>> + }
>> +
>> + /* clk name */
>> + of_property_read_string(np, "clock-output-names", &clk_name);
>> +
>> + /* divider clock register */
>> + clk = clk_register_divider(NULL, clk_name, parent_name,
>> + 0, pic32_clk_regbase,
>> + OSC_FRCDIV_SHIFT, OSC_FRCDIV_MASK,
>> + CLK_DIVIDER_POWER_OF_TWO, &lock);
>> +
>> + if (IS_ERR_OR_NULL(clk)) {
> Use IS_ERR(), not IS_ERR_OR_NULL(). NULL is a valid clock
> pointer.

ack.

>> + pr_err("frcdiv_clk not registered\n");
>> + return;
>> + }
>> +
>> + pic32_of_clk_register_clkdev(np, clk);
>> +}
>> +
>> +static const struct of_device_id pic32_clk_match[] __initconst = {
>> + {
>> + .compatible = "microchip,pic32mzda-refoclk",
>> + .data = of_refo_clk_setup,
>> + },
>> + {
>> + .compatible = "microchip,pic32mzda-pbclk",
>> + .data = of_periph_clk_setup,
>> + },
>> + {
>> + .compatible = "microchip,pic32mzda-syspll",
>> + .data = of_sys_pll_setup,
>> + },
>> + {
>> + .compatible = "microchip,pic32mzda-sosc",
>> + .data = of_sosc_clk_setup,
>> + },
>> + {
>> + .compatible = "microchip,pic32mzda-frcdivclk",
>> + .data = of_frcdiv_setup,
>> + },
>> + {
>> + .compatible = "microchip,pic32mzda-sysclk-v2",
>> + .data = of_sys_mux_slew_setup,
>> + },
>> + {}
>> +};
>> +
>> +static irqreturn_t pic32_fscm_isr_handler(int irq, void *data)
>> +{
>> + u32 v = clk_readl(pic32_clk_regbase);
>> +
>> + if (v & OSC_CLK_FAILED)
>> + pr_info("pic32-clk: FSCM detected clk failure.\n");
> Why isn't this pr_err()?

ack

>> +
>> + return IRQ_HANDLED;
>> +}
>> +
>> +static int pic32_fscm_nmi(struct notifier_block *nb,
>> + unsigned long action, void *data)
>> +{
>> + pic32_fscm_isr_handler(0, NULL);
>> + return NOTIFY_OK;
>> +}
>> +
>> +static struct notifier_block failsafe_clk_notifier = {
>> + .notifier_call = pic32_fscm_nmi,
>> +};
>> +
>> +static void __init of_pic32_soc_clock_init(struct device_node *np)
>> +{
>> + int ret, nmi = 0, irq;
>> + struct resource r;
>> + struct device_node *childnp;
>> + const struct of_device_id *clk_id;
>> + void (*clk_setup)(struct device_node *);
>> +
>> + if (of_address_to_resource(np, 0, &r))
>> + panic("Failed to get clk-pll memory region\n");
>> +
>> + if (!request_mem_region(r.start, resource_size(&r), r.name))
>> + panic("%s: request_region failed\n", np->name);
>> +
>> + pic32_clk_regbase = ioremap_nocache(r.start, resource_size(&r));
> Can we use ioremap() instead? If so, we could use
> of_io_request_and_map() then.

ack. Will replace with of_io_request_and_map().

>> + if (!pic32_clk_regbase)
>> + panic("pic32-clk: failed to map registers\n");
>> +
>> + irq = irq_of_parse_and_map(np, 0);
>> + if (!irq) {
>> + pr_warn("pic32-clk: irq not provided for FSCM; use nmi.\n");
>> + nmi = 1;
>> + }
> Why can't we parse and map the irq down there at the if (nmi)?
> Lose the local variable.

ack.

>> +
>> + for_each_child_of_node(np, childnp) {
>> + clk_id = of_match_node(pic32_clk_match, childnp);
>> + if (!clk_id)
>> + continue;
>> + clk_setup = clk_id->data;
>> + clk_setup(childnp);
>> + }
> Finally, can this be a platform driver instead of using
> OF_CLK_DECLARE()?
>
I'll prefer to use OF_CLK_DECLARE() way. please provide some reference otherwise.