clk: add Dove PLL divider support for GPU, VMeta and AXI clocks

Add support for the Dove PLL dividers, which are used to generate the
clocks for the AXI bus, as well as the GPU and VMeta peripherals.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Michael Turquette <mturquette@baylibre.com>

drivers/clk/mvebu/Makefile

@@ -7,6 +7,6 @@ obj-$(CONFIG_ARMADA_375_CLK)	+= armada-375.o
 obj-$(CONFIG_ARMADA_38X_CLK)	+= armada-38x.o
 obj-$(CONFIG_ARMADA_39X_CLK)	+= armada-39x.o
 obj-$(CONFIG_ARMADA_XP_CLK)	+= armada-xp.o
-obj-$(CONFIG_DOVE_CLK)		+= dove.o
+obj-$(CONFIG_DOVE_CLK)		+= dove.o dove-divider.o
 obj-$(CONFIG_KIRKWOOD_CLK)	+= kirkwood.o
 obj-$(CONFIG_ORION_CLK)		+= orion.o

drivers/clk/mvebu/dove-divider.c

@@ -0,0 +1,262 @@
+/*
+ * Marvell Dove PMU Core PLL divider driver
+ *
+ * Cleaned up by substantially rewriting, and converted to DT by
+ * Russell King.  Origin is not known.
+ */
+#include <linux/clk-provider.h>
+#include <linux/delay.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+
+#include "dove-divider.h"
+
+struct dove_clk {
+	const char *name;
+	struct clk_hw hw;
+	void __iomem *base;
+	spinlock_t *lock;
+	u8 div_bit_start;
+	u8 div_bit_end;
+	u8 div_bit_load;
+	u8 div_bit_size;
+	u32 *divider_table;
+};
+
+enum {
+	DIV_CTRL0 = 0,
+	DIV_CTRL1 = 4,
+	DIV_CTRL1_N_RESET_MASK = BIT(10),
+};
+
+#define to_dove_clk(hw) container_of(hw, struct dove_clk, hw)
+
+static void dove_load_divider(void __iomem *base, u32 val, u32 mask, u32 load)
+{
+	u32 v;
+
+	v = readl_relaxed(base + DIV_CTRL1) | DIV_CTRL1_N_RESET_MASK;
+	writel_relaxed(v, base + DIV_CTRL1);
+
+	v = (readl_relaxed(base + DIV_CTRL0) & ~(mask | load)) | val;
+	writel_relaxed(v, base + DIV_CTRL0);
+	writel_relaxed(v | load, base + DIV_CTRL0);
+	ndelay(250);
+	writel_relaxed(v, base + DIV_CTRL0);
+}
+
+static unsigned int dove_get_divider(struct dove_clk *dc)
+{
+	unsigned int divider;
+	u32 val;
+
+	val = readl_relaxed(dc->base + DIV_CTRL0);
+	val >>= dc->div_bit_start;
+
+	divider = val & ~(~0 << dc->div_bit_size);
+
+	if (dc->divider_table)
+		divider = dc->divider_table[divider];
+
+	return divider;
+}
+
+static int dove_calc_divider(const struct dove_clk *dc, unsigned long rate,
+			     unsigned long parent_rate, bool set)
+{
+	unsigned int divider, max;
+
+	divider = DIV_ROUND_CLOSEST(parent_rate, rate);
+
+	if (dc->divider_table) {
+		unsigned int i;
+
+		for (i = 0; dc->divider_table[i]; i++)
+			if (divider == dc->divider_table[i]) {
+				divider = i;
+				break;
+			}
+
+		if (!dc->divider_table[i])
+			return -EINVAL;
+	} else {
+		max = 1 << dc->div_bit_size;
+
+		if (set && (divider == 0 || divider >= max))
+			return -EINVAL;
+		if (divider >= max)
+			divider = max - 1;
+		else if (divider == 0)
+			divider = 1;
+	}
+
+	return divider;
+}
+
+static unsigned long dove_recalc_rate(struct clk_hw *hw, unsigned long parent)
+{
+	struct dove_clk *dc = to_dove_clk(hw);
+	unsigned int divider = dove_get_divider(dc);
+	unsigned long rate = DIV_ROUND_CLOSEST(parent, divider);
+
+	pr_debug("%s(): %s divider=%u parent=%lu rate=%lu\n",
+		 __func__, dc->name, divider, parent, rate);
+
+	return rate;
+}
+
+static long dove_round_rate(struct clk_hw *hw, unsigned long rate,
+			    unsigned long *parent)
+{
+	struct dove_clk *dc = to_dove_clk(hw);
+	unsigned long parent_rate = *parent;
+	int divider;
+
+	divider = dove_calc_divider(dc, rate, parent_rate, false);
+	if (divider < 0)
+		return divider;
+
+	rate = DIV_ROUND_CLOSEST(parent_rate, divider);
+
+	pr_debug("%s(): %s divider=%u parent=%lu rate=%lu\n",
+		 __func__, dc->name, divider, parent_rate, rate);
+
+	return rate;
+}
+
+static int dove_set_clock(struct clk_hw *hw, unsigned long rate,
+			  unsigned long parent_rate)
+{
+	struct dove_clk *dc = to_dove_clk(hw);
+	u32 mask, load, div;
+	int divider;
+
+	divider = dove_calc_divider(dc, rate, parent_rate, true);
+	if (divider < 0)
+		return divider;
+
+	pr_debug("%s(): %s divider=%u parent=%lu rate=%lu\n",
+		 __func__, dc->name, divider, parent_rate, rate);
+
+	div = (u32)divider << dc->div_bit_start;
+	mask = ~(~0 << dc->div_bit_size) << dc->div_bit_start;
+	load = BIT(dc->div_bit_load);
+
+	spin_lock(dc->lock);
+	dove_load_divider(dc->base, div, mask, load);
+	spin_unlock(dc->lock);
+
+	return 0;
+}
+
+static const struct clk_ops dove_divider_ops = {
+	.set_rate	= dove_set_clock,
+	.round_rate	= dove_round_rate,
+	.recalc_rate	= dove_recalc_rate,
+};
+
+static struct clk *clk_register_dove_divider(struct device *dev,
+	struct dove_clk *dc, const char **parent_names, size_t num_parents,
+	void __iomem *base)
+{
+	char name[32];
+	struct clk_init_data init = {
+		.name = name,
+		.ops = &dove_divider_ops,
+		.parent_names = parent_names,
+		.num_parents = num_parents,
+	};
+
+	strlcpy(name, dc->name, sizeof(name));
+
+	dc->hw.init = &init;
+	dc->base = base;
+	dc->div_bit_size = dc->div_bit_end - dc->div_bit_start + 1;
+
+	return clk_register(dev, &dc->hw);
+}
+
+static DEFINE_SPINLOCK(dove_divider_lock);
+
+static u32 axi_divider[] = {-1, 2, 1, 3, 4, 6, 5, 7, 8, 10, 9, 0};
+
+static struct dove_clk dove_hw_clocks[4] = {
+	{
+		.name = "axi",
+		.lock = &dove_divider_lock,
+		.div_bit_start = 1,
+		.div_bit_end = 6,
+		.div_bit_load = 7,
+		.divider_table = axi_divider,
+	}, {
+		.name = "gpu",
+		.lock = &dove_divider_lock,
+		.div_bit_start = 8,
+		.div_bit_end = 13,
+		.div_bit_load = 14,
+	}, {
+		.name = "vmeta",
+		.lock = &dove_divider_lock,
+		.div_bit_start = 15,
+		.div_bit_end = 20,
+		.div_bit_load = 21,
+	}, {
+		.name = "lcd",
+		.lock = &dove_divider_lock,
+		.div_bit_start = 22,
+		.div_bit_end = 27,
+		.div_bit_load = 28,
+	},
+};
+
+static const char *core_pll[] = {
+	"core-pll",
+};
+
+static int dove_divider_init(struct device *dev, void __iomem *base,
+	struct clk **clks)
+{
+	struct clk *clk;
+	int i;
+
+	/*
+	 * Create the core PLL clock.  We treat this as a fixed rate
+	 * clock as we don't know any better, and documentation is sparse.
+	 */
+	clk = clk_register_fixed_rate(dev, core_pll[0], NULL, CLK_IS_ROOT,
+				      2000000000UL);
+	if (IS_ERR(clk))
+		return PTR_ERR(clk);
+
+	for (i = 0; i < ARRAY_SIZE(dove_hw_clocks); i++)
+		clks[i] = clk_register_dove_divider(dev, &dove_hw_clocks[i],
+						    core_pll,
+						    ARRAY_SIZE(core_pll), base);
+
+	return 0;
+}
+
+static struct clk *dove_divider_clocks[4];
+
+static struct clk_onecell_data dove_divider_data = {
+	.clks = dove_divider_clocks,
+	.clk_num = ARRAY_SIZE(dove_divider_clocks),
+};
+
+void __init dove_divider_clk_init(struct device_node *np)
+{
+	void *base;
+
+	base = of_iomap(np, 0);
+	if (WARN_ON(!base))
+		return;
+
+	if (WARN_ON(dove_divider_init(NULL, base, dove_divider_clocks))) {
+		iounmap(base);
+		return;
+	}
+
+	of_clk_add_provider(np, of_clk_src_onecell_get, &dove_divider_data);
+}

drivers/clk/mvebu/dove-divider.h

@@ -0,0 +1,6 @@
+#ifndef DOVE_DIVIDER_H
+#define DOVE_DIVIDER_H
+
+void __init dove_divider_clk_init(struct device_node *np);
+
+#endif

drivers/clk/mvebu/dove.c

@@ -17,6 +17,7 @@
 #include <linux/io.h>
 #include <linux/of.h>
 #include "common.h"
+#include "dove-divider.h"
 
 /*
  * Core Clocks
@@ -184,9 +185,14 @@ static void __init dove_clk_init(struct device_node *np)
 {
 	struct device_node *cgnp =
 		of_find_compatible_node(NULL, NULL, "marvell,dove-gating-clock");
+	struct device_node *ddnp =
+		of_find_compatible_node(NULL, NULL, "marvell,dove-divider-clock");
 
 	mvebu_coreclk_setup(np, &dove_coreclks);
 
+	if (ddnp)
+		dove_divider_clk_init(ddnp);
+
 	if (cgnp)
 		mvebu_clk_gating_setup(cgnp, dove_gating_desc);
 }