Merge branch 'armsoc/for-3.15/drivers' of https://github.com/broadcom/mach-bcm into wip/llct-capri-support-1

8 files changed, 2642 insertions, 0 deletions

diff --git a/drivers/clk/Kconfig b/drivers/clk/Kconfig
index 7641965d208..f9f605695e4 100644
--- a/drivers/clk/Kconfig
+++ b/drivers/clk/Kconfig
@@ -111,4 +111,5 @@ source "drivers/clk/qcom/Kconfig"
 
 endmenu
 
+source "drivers/clk/bcm/Kconfig"
 source "drivers/clk/mvebu/Kconfig"
diff --git a/drivers/clk/Makefile b/drivers/clk/Makefile
index a367a983171..88af4a399d6 100644
--- a/drivers/clk/Makefile
+++ b/drivers/clk/Makefile
@@ -29,6 +29,7 @@ obj-$(CONFIG_ARCH_VT8500)		+= clk-vt8500.o
 obj-$(CONFIG_COMMON_CLK_WM831X)		+= clk-wm831x.o
 obj-$(CONFIG_COMMON_CLK_XGENE)		+= clk-xgene.o
 obj-$(CONFIG_COMMON_CLK_AT91)		+= at91/
+obj-$(CONFIG_ARCH_BCM_MOBILE)		+= bcm/
 obj-$(CONFIG_ARCH_HI3xxx)		+= hisilicon/
 obj-$(CONFIG_COMMON_CLK_KEYSTONE)	+= keystone/
 ifeq ($(CONFIG_COMMON_CLK), y)
diff --git a/drivers/clk/bcm/Kconfig b/drivers/clk/bcm/Kconfig
new file mode 100644
index 00000000000..a7262fb8ce5
--- /dev/null
+++ b/drivers/clk/bcm/Kconfig
@@ -0,0 +1,9 @@
+config CLK_BCM_KONA
+	bool "Broadcom Kona CCU clock support"
+	depends on ARCH_BCM_MOBILE
+	depends on COMMON_CLK
+	default y
+	help
+	  Enable common clock framework support for Broadcom SoCs
+	  using "Kona" style clock control units, including those
+	  in the BCM281xx family.
diff --git a/drivers/clk/bcm/Makefile b/drivers/clk/bcm/Makefile
new file mode 100644
index 00000000000..cf93359aa86
--- /dev/null
+++ b/drivers/clk/bcm/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_CLK_BCM_KONA)	+= clk-kona.o
+obj-$(CONFIG_CLK_BCM_KONA)	+= clk-kona-setup.o
+obj-$(CONFIG_CLK_BCM_KONA)	+= clk-bcm281xx.o
diff --git a/drivers/clk/bcm/clk-bcm281xx.c b/drivers/clk/bcm/clk-bcm281xx.c
new file mode 100644
index 00000000000..3c66de696ae
--- /dev/null
+++ b/drivers/clk/bcm/clk-bcm281xx.c
@@ -0,0 +1,416 @@
+/*
+ * Copyright (C) 2013 Broadcom Corporation
+ * Copyright 2013 Linaro Limited
+ *
+ * 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.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include "clk-kona.h"
+#include "dt-bindings/clock/bcm281xx.h"
+
+/* bcm11351 CCU device tree "compatible" strings */
+#define BCM11351_DT_ROOT_CCU_COMPAT	"brcm,bcm11351-root-ccu"
+#define BCM11351_DT_AON_CCU_COMPAT	"brcm,bcm11351-aon-ccu"
+#define BCM11351_DT_HUB_CCU_COMPAT	"brcm,bcm11351-hub-ccu"
+#define BCM11351_DT_MASTER_CCU_COMPAT	"brcm,bcm11351-master-ccu"
+#define BCM11351_DT_SLAVE_CCU_COMPAT	"brcm,bcm11351-slave-ccu"
+
+/* Root CCU clocks */
+
+static struct peri_clk_data frac_1m_data = {
+	.gate		= HW_SW_GATE(0x214, 16, 0, 1),
+	.trig		= TRIGGER(0x0e04, 0),
+	.div		= FRAC_DIVIDER(0x0e00, 0, 22, 16),
+	.clocks		= CLOCKS("ref_crystal"),
+};
+
+/* AON CCU clocks */
+
+static struct peri_clk_data hub_timer_data = {
+	.gate		= HW_SW_GATE(0x0414, 16, 0, 1),
+	.clocks		= CLOCKS("bbl_32k",
+				 "frac_1m",
+				 "dft_19_5m"),
+	.sel		= SELECTOR(0x0a10, 0, 2),
+	.trig		= TRIGGER(0x0a40, 4),
+};
+
+static struct peri_clk_data pmu_bsc_data = {
+	.gate		= HW_SW_GATE(0x0418, 16, 0, 1),
+	.clocks		= CLOCKS("ref_crystal",
+				 "pmu_bsc_var",
+				 "bbl_32k"),
+	.sel		= SELECTOR(0x0a04, 0, 2),
+	.div		= DIVIDER(0x0a04, 3, 4),
+	.trig		= TRIGGER(0x0a40, 0),
+};
+
+static struct peri_clk_data pmu_bsc_var_data = {
+	.clocks		= CLOCKS("var_312m",
+				 "ref_312m"),
+	.sel		= SELECTOR(0x0a00, 0, 2),
+	.div		= DIVIDER(0x0a00, 4, 5),
+	.trig		= TRIGGER(0x0a40, 2),
+};
+
+/* Hub CCU clocks */
+
+static struct peri_clk_data tmon_1m_data = {
+	.gate		= HW_SW_GATE(0x04a4, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "frac_1m"),
+	.sel		= SELECTOR(0x0e74, 0, 2),
+	.trig		= TRIGGER(0x0e84, 1),
+};
+
+/* Master CCU clocks */
+
+static struct peri_clk_data sdio1_data = {
+	.gate		= HW_SW_GATE(0x0358, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_52m",
+				 "ref_52m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a28, 0, 3),
+	.div		= DIVIDER(0x0a28, 4, 14),
+	.trig		= TRIGGER(0x0afc, 9),
+};
+
+static struct peri_clk_data sdio2_data = {
+	.gate		= HW_SW_GATE(0x035c, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_52m",
+				 "ref_52m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a2c, 0, 3),
+	.div		= DIVIDER(0x0a2c, 4, 14),
+	.trig		= TRIGGER(0x0afc, 10),
+};
+
+static struct peri_clk_data sdio3_data = {
+	.gate		= HW_SW_GATE(0x0364, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_52m",
+				 "ref_52m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a34, 0, 3),
+	.div		= DIVIDER(0x0a34, 4, 14),
+	.trig		= TRIGGER(0x0afc, 12),
+};
+
+static struct peri_clk_data sdio4_data = {
+	.gate		= HW_SW_GATE(0x0360, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_52m",
+				 "ref_52m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a30, 0, 3),
+	.div		= DIVIDER(0x0a30, 4, 14),
+	.trig		= TRIGGER(0x0afc, 11),
+};
+
+static struct peri_clk_data usb_ic_data = {
+	.gate		= HW_SW_GATE(0x0354, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_96m",
+				 "ref_96m"),
+	.div		= FIXED_DIVIDER(2),
+	.sel		= SELECTOR(0x0a24, 0, 2),
+	.trig		= TRIGGER(0x0afc, 7),
+};
+
+/* also called usbh_48m */
+static struct peri_clk_data hsic2_48m_data = {
+	.gate		= HW_SW_GATE(0x0370, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a38, 0, 2),
+	.div		= FIXED_DIVIDER(2),
+	.trig		= TRIGGER(0x0afc, 5),
+};
+
+/* also called usbh_12m */
+static struct peri_clk_data hsic2_12m_data = {
+	.gate		= HW_SW_GATE(0x0370, 20, 4, 5),
+	.div		= DIVIDER(0x0a38, 12, 2),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_96m",
+				 "ref_96m"),
+	.pre_div	= FIXED_DIVIDER(2),
+	.sel		= SELECTOR(0x0a38, 0, 2),
+	.trig		= TRIGGER(0x0afc, 5),
+};
+
+/* Slave CCU clocks */
+
+static struct peri_clk_data uartb_data = {
+	.gate		= HW_SW_GATE(0x0400, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_156m",
+				 "ref_156m"),
+	.sel		= SELECTOR(0x0a10, 0, 2),
+	.div		= FRAC_DIVIDER(0x0a10, 4, 12, 8),
+	.trig		= TRIGGER(0x0afc, 2),
+};
+
+static struct peri_clk_data uartb2_data = {
+	.gate		= HW_SW_GATE(0x0404, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_156m",
+				 "ref_156m"),
+	.sel		= SELECTOR(0x0a14, 0, 2),
+	.div		= FRAC_DIVIDER(0x0a14, 4, 12, 8),
+	.trig		= TRIGGER(0x0afc, 3),
+};
+
+static struct peri_clk_data uartb3_data = {
+	.gate		= HW_SW_GATE(0x0408, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_156m",
+				 "ref_156m"),
+	.sel		= SELECTOR(0x0a18, 0, 2),
+	.div		= FRAC_DIVIDER(0x0a18, 4, 12, 8),
+	.trig		= TRIGGER(0x0afc, 4),
+};
+
+static struct peri_clk_data uartb4_data = {
+	.gate		= HW_SW_GATE(0x0408, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_156m",
+				 "ref_156m"),
+	.sel		= SELECTOR(0x0a1c, 0, 2),
+	.div		= FRAC_DIVIDER(0x0a1c, 4, 12, 8),
+	.trig		= TRIGGER(0x0afc, 5),
+};
+
+static struct peri_clk_data ssp0_data = {
+	.gate		= HW_SW_GATE(0x0410, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_104m",
+				 "ref_104m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a20, 0, 3),
+	.div		= DIVIDER(0x0a20, 4, 14),
+	.trig		= TRIGGER(0x0afc, 6),
+};
+
+static struct peri_clk_data ssp2_data = {
+	.gate		= HW_SW_GATE(0x0418, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_104m",
+				 "ref_104m",
+				 "var_96m",
+				 "ref_96m"),
+	.sel		= SELECTOR(0x0a28, 0, 3),
+	.div		= DIVIDER(0x0a28, 4, 14),
+	.trig		= TRIGGER(0x0afc, 8),
+};
+
+static struct peri_clk_data bsc1_data = {
+	.gate		= HW_SW_GATE(0x0458, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_104m",
+				 "ref_104m",
+				 "var_13m",
+				 "ref_13m"),
+	.sel		= SELECTOR(0x0a64, 0, 3),
+	.trig		= TRIGGER(0x0afc, 23),
+};
+
+static struct peri_clk_data bsc2_data = {
+	.gate		= HW_SW_GATE(0x045c, 18, 2, 3),
+	.clocks	= CLOCKS("ref_crystal",
+				 "var_104m",
+				 "ref_104m",
+				 "var_13m",
+				 "ref_13m"),
+	.sel		= SELECTOR(0x0a68, 0, 3),
+	.trig		= TRIGGER(0x0afc, 24),
+};
+
+static struct peri_clk_data bsc3_data = {
+	.gate		= HW_SW_GATE(0x0484, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_104m",
+				 "ref_104m",
+				 "var_13m",
+				 "ref_13m"),
+	.sel		= SELECTOR(0x0a84, 0, 3),
+	.trig		= TRIGGER(0x0b00, 2),
+};
+
+static struct peri_clk_data pwm_data = {
+	.gate		= HW_SW_GATE(0x0468, 18, 2, 3),
+	.clocks		= CLOCKS("ref_crystal",
+				 "var_104m"),
+	.sel		= SELECTOR(0x0a70, 0, 2),
+	.div		= DIVIDER(0x0a70, 4, 3),
+	.trig		= TRIGGER(0x0afc, 15),
+};
+
+/*
+ * CCU setup routines
+ *
+ * These are called from kona_dt_ccu_setup() to initialize the array
+ * of clocks provided by the CCU.  Once allocated, the entries in
+ * the array are initialized by calling kona_clk_setup() with the
+ * initialization data for each clock.  They return 0 if successful
+ * or an error code otherwise.
+ */
+static int __init bcm281xx_root_ccu_clks_setup(struct ccu_data *ccu)
+{
+	struct clk **clks;
+	size_t count = BCM281XX_ROOT_CCU_CLOCK_COUNT;
+
+	clks = kzalloc(count * sizeof(*clks), GFP_KERNEL);
+	if (!clks) {
+		pr_err("%s: failed to allocate root clocks\n", __func__);
+		return -ENOMEM;
+	}
+	ccu->data.clks = clks;
+	ccu->data.clk_num = count;
+
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_ROOT_CCU_FRAC_1M, frac_1m);
+
+	return 0;
+}
+
+static int __init bcm281xx_aon_ccu_clks_setup(struct ccu_data *ccu)
+{
+	struct clk **clks;
+	size_t count = BCM281XX_AON_CCU_CLOCK_COUNT;
+
+	clks = kzalloc(count * sizeof(*clks), GFP_KERNEL);
+	if (!clks) {
+		pr_err("%s: failed to allocate aon clocks\n", __func__);
+		return -ENOMEM;
+	}
+	ccu->data.clks = clks;
+	ccu->data.clk_num = count;
+
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_AON_CCU_HUB_TIMER, hub_timer);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_AON_CCU_PMU_BSC, pmu_bsc);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_AON_CCU_PMU_BSC_VAR, pmu_bsc_var);
+
+	return 0;
+}
+
+static int __init bcm281xx_hub_ccu_clks_setup(struct ccu_data *ccu)
+{
+	struct clk **clks;
+	size_t count = BCM281XX_HUB_CCU_CLOCK_COUNT;
+
+	clks = kzalloc(count * sizeof(*clks), GFP_KERNEL);
+	if (!clks) {
+		pr_err("%s: failed to allocate hub clocks\n", __func__);
+		return -ENOMEM;
+	}
+	ccu->data.clks = clks;
+	ccu->data.clk_num = count;
+
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_HUB_CCU_TMON_1M, tmon_1m);
+
+	return 0;
+}
+
+static int __init bcm281xx_master_ccu_clks_setup(struct ccu_data *ccu)
+{
+	struct clk **clks;
+	size_t count = BCM281XX_MASTER_CCU_CLOCK_COUNT;
+
+	clks = kzalloc(count * sizeof(*clks), GFP_KERNEL);
+	if (!clks) {
+		pr_err("%s: failed to allocate master clocks\n", __func__);
+		return -ENOMEM;
+	}
+	ccu->data.clks = clks;
+	ccu->data.clk_num = count;
+
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_SDIO1, sdio1);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_SDIO2, sdio2);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_SDIO3, sdio3);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_SDIO4, sdio4);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_USB_IC, usb_ic);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_HSIC2_48M, hsic2_48m);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_MASTER_CCU_HSIC2_12M, hsic2_12m);
+
+	return 0;
+}
+
+static int __init bcm281xx_slave_ccu_clks_setup(struct ccu_data *ccu)
+{
+	struct clk **clks;
+	size_t count = BCM281XX_SLAVE_CCU_CLOCK_COUNT;
+
+	clks = kzalloc(count * sizeof(*clks), GFP_KERNEL);
+	if (!clks) {
+		pr_err("%s: failed to allocate slave clocks\n", __func__);
+		return -ENOMEM;
+	}
+	ccu->data.clks = clks;
+	ccu->data.clk_num = count;
+
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_UARTB, uartb);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_UARTB2, uartb2);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_UARTB3, uartb3);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_UARTB4, uartb4);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_SSP0, ssp0);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_SSP2, ssp2);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_BSC1, bsc1);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_BSC2, bsc2);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_BSC3, bsc3);
+	PERI_CLK_SETUP(clks, ccu, BCM281XX_SLAVE_CCU_PWM, pwm);
+
+	return 0;
+}
+
+/* Device tree match table callback functions */
+
+static void __init kona_dt_root_ccu_setup(struct device_node *node)
+{
+	kona_dt_ccu_setup(node, bcm281xx_root_ccu_clks_setup);
+}
+
+static void __init kona_dt_aon_ccu_setup(struct device_node *node)
+{
+	kona_dt_ccu_setup(node, bcm281xx_aon_ccu_clks_setup);
+}
+
+static void __init kona_dt_hub_ccu_setup(struct device_node *node)
+{
+	kona_dt_ccu_setup(node, bcm281xx_hub_ccu_clks_setup);
+}
+
+static void __init kona_dt_master_ccu_setup(struct device_node *node)
+{
+	kona_dt_ccu_setup(node, bcm281xx_master_ccu_clks_setup);
+}
+
+static void __init kona_dt_slave_ccu_setup(struct device_node *node)
+{
+	kona_dt_ccu_setup(node, bcm281xx_slave_ccu_clks_setup);
+}
+
+CLK_OF_DECLARE(bcm11351_root_ccu, BCM11351_DT_ROOT_CCU_COMPAT,
+			kona_dt_root_ccu_setup);
+CLK_OF_DECLARE(bcm11351_aon_ccu, BCM11351_DT_AON_CCU_COMPAT,
+			kona_dt_aon_ccu_setup);
+CLK_OF_DECLARE(bcm11351_hub_ccu, BCM11351_DT_HUB_CCU_COMPAT,
+			kona_dt_hub_ccu_setup);
+CLK_OF_DECLARE(bcm11351_master_ccu, BCM11351_DT_MASTER_CCU_COMPAT,
+			kona_dt_master_ccu_setup);
+CLK_OF_DECLARE(bcm11351_slave_ccu, BCM11351_DT_SLAVE_CCU_COMPAT,
+			kona_dt_slave_ccu_setup);
diff --git a/drivers/clk/bcm/clk-kona-setup.c b/drivers/clk/bcm/clk-kona-setup.c
new file mode 100644
index 00000000000..c7607feb18d
--- /dev/null
+++ b/drivers/clk/bcm/clk-kona-setup.c
@@ -0,0 +1,769 @@
+/*
+ * Copyright (C) 2013 Broadcom Corporation
+ * Copyright 2013 Linaro Limited
+ *
+ * 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.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/io.h>
+#include <linux/of_address.h>
+
+#include "clk-kona.h"
+
+/* These are used when a selector or trigger is found to be unneeded */
+#define selector_clear_exists(sel)	((sel)->width = 0)
+#define trigger_clear_exists(trig)	FLAG_CLEAR(trig, TRIG, EXISTS)
+
+LIST_HEAD(ccu_list);	/* The list of set up CCUs */
+
+/* Validity checking */
+
+static bool clk_requires_trigger(struct kona_clk *bcm_clk)
+{
+	struct peri_clk_data *peri = bcm_clk->peri;
+	struct bcm_clk_sel *sel;
+	struct bcm_clk_div *div;
+
+	if (bcm_clk->type != bcm_clk_peri)
+		return false;
+
+	sel = &peri->sel;
+	if (sel->parent_count && selector_exists(sel))
+		return true;
+
+	div = &peri->div;
+	if (!divider_exists(div))
+		return false;
+
+	/* Fixed dividers don't need triggers */
+	if (!divider_is_fixed(div))
+		return true;
+
+	div = &peri->pre_div;
+
+	return divider_exists(div) && !divider_is_fixed(div);
+}
+
+static bool peri_clk_data_offsets_valid(struct kona_clk *bcm_clk)
+{
+	struct peri_clk_data *peri;
+	struct bcm_clk_gate *gate;
+	struct bcm_clk_div *div;
+	struct bcm_clk_sel *sel;
+	struct bcm_clk_trig *trig;
+	const char *name;
+	u32 range;
+	u32 limit;
+
+	BUG_ON(bcm_clk->type != bcm_clk_peri);
+	peri = bcm_clk->peri;
+	name = bcm_clk->name;
+	range = bcm_clk->ccu->range;
+
+	limit = range - sizeof(u32);
+	limit = round_down(limit, sizeof(u32));
+
+	gate = &peri->gate;
+	if (gate_exists(gate)) {
+		if (gate->offset > limit) {
+			pr_err("%s: bad gate offset for %s (%u > %u)\n",
+				__func__, name, gate->offset, limit);
+			return false;
+		}
+	}
+
+	div = &peri->div;
+	if (divider_exists(div)) {
+		if (div->offset > limit) {
+			pr_err("%s: bad divider offset for %s (%u > %u)\n",
+				__func__, name, div->offset, limit);
+			return false;
+		}
+	}
+
+	div = &peri->pre_div;
+	if (divider_exists(div)) {
+		if (div->offset > limit) {
+			pr_err("%s: bad pre-divider offset for %s "
+					"(%u > %u)\n",
+				__func__, name, div->offset, limit);
+			return false;
+		}
+	}
+
+	sel = &peri->sel;
+	if (selector_exists(sel)) {
+		if (sel->offset > limit) {
+			pr_err("%s: bad selector offset for %s (%u > %u)\n",
+				__func__, name, sel->offset, limit);
+			return false;
+		}
+	}
+
+	trig = &peri->trig;
+	if (trigger_exists(trig)) {
+		if (trig->offset > limit) {
+			pr_err("%s: bad trigger offset for %s (%u > %u)\n",
+				__func__, name, trig->offset, limit);
+			return false;
+		}
+	}
+
+	trig = &peri->pre_trig;
+	if (trigger_exists(trig)) {
+		if (trig->offset > limit) {
+			pr_err("%s: bad pre-trigger offset for %s (%u > %u)\n",
+				__func__, name, trig->offset, limit);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+/* A bit position must be less than the number of bits in a 32-bit register. */
+static bool bit_posn_valid(u32 bit_posn, const char *field_name,
+			const char *clock_name)
+{
+	u32 limit = BITS_PER_BYTE * sizeof(u32) - 1;
+
+	if (bit_posn > limit) {
+		pr_err("%s: bad %s bit for %s (%u > %u)\n", __func__,
+			field_name, clock_name, bit_posn, limit);
+		return false;
+	}
+	return true;
+}
+
+/*
+ * A bitfield must be at least 1 bit wide.  Both the low-order and
+ * high-order bits must lie within a 32-bit register.  We require
+ * fields to be less than 32 bits wide, mainly because we use
+ * shifting to produce field masks, and shifting a full word width
+ * is not well-defined by the C standard.
+ */
+static bool bitfield_valid(u32 shift, u32 width, const char *field_name,
+			const char *clock_name)
+{
+	u32 limit = BITS_PER_BYTE * sizeof(u32);
+
+	if (!width) {
+		pr_err("%s: bad %s field width 0 for %s\n", __func__,
+			field_name, clock_name);
+		return false;
+	}
+	if (shift + width > limit) {
+		pr_err("%s: bad %s for %s (%u + %u > %u)\n", __func__,
+			field_name, clock_name, shift, width, limit);
+		return false;
+	}
+	return true;
+}
+
+/*
+ * All gates, if defined, have a status bit, and for hardware-only
+ * gates, that's it.  Gates that can be software controlled also
+ * have an enable bit.  And a gate that can be hardware or software
+ * controlled will have a hardware/software select bit.
+ */
+static bool gate_valid(struct bcm_clk_gate *gate, const char *field_name,
+			const char *clock_name)
+{
+	if (!bit_posn_valid(gate->status_bit, "gate status", clock_name))
+		return false;
+
+	if (gate_is_sw_controllable(gate)) {
+		if (!bit_posn_valid(gate->en_bit, "gate enable", clock_name))
+			return false;
+
+		if (gate_is_hw_controllable(gate)) {
+			if (!bit_posn_valid(gate->hw_sw_sel_bit,
+						"gate hw/sw select",
+						clock_name))
+				return false;
+		}
+	} else {
+		BUG_ON(!gate_is_hw_controllable(gate));
+	}
+
+	return true;
+}
+
+/*
+ * A selector bitfield must be valid.  Its parent_sel array must
+ * also be reasonable for the field.
+ */
+static bool sel_valid(struct bcm_clk_sel *sel, const char *field_name,
+			const char *clock_name)
+{
+	if (!bitfield_valid(sel->shift, sel->width, field_name, clock_name))
+		return false;
+
+	if (sel->parent_count) {
+		u32 max_sel;
+		u32 limit;
+
+		/*
+		 * Make sure the selector field can hold all the
+		 * selector values we expect to be able to use.  A
+		 * clock only needs to have a selector defined if it
+		 * has more than one parent.  And in that case the
+		 * highest selector value will be in the last entry
+		 * in the array.
+		 */
+		max_sel = sel->parent_sel[sel->parent_count - 1];
+		limit = (1 << sel->width) - 1;
+		if (max_sel > limit) {
+			pr_err("%s: bad selector for %s "
+					"(%u needs > %u bits)\n",
+				__func__, clock_name, max_sel,
+				sel->width);
+			return false;
+		}
+	} else {
+		pr_warn("%s: ignoring selector for %s (no parents)\n",
+			__func__, clock_name);
+		selector_clear_exists(sel);
+		kfree(sel->parent_sel);
+		sel->parent_sel = NULL;
+	}
+
+	return true;
+}
+
+/*
+ * A fixed divider just needs to be non-zero.  A variable divider
+ * has to have a valid divider bitfield, and if it has a fraction,
+ * the width of the fraction must not be no more than the width of
+ * the divider as a whole.
+ */
+static bool div_valid(struct bcm_clk_div *div, const char *field_name,
+			const char *clock_name)
+{
+	if (divider_is_fixed(div)) {
+		/* Any fixed divider value but 0 is OK */
+		if (div->fixed == 0) {
+			pr_err("%s: bad %s fixed value 0 for %s\n", __func__,
+				field_name, clock_name);
+			return false;
+		}
+		return true;
+	}
+	if (!bitfield_valid(div->shift, div->width, field_name, clock_name))
+		return false;
+
+	if (divider_has_fraction(div))
+		if (div->frac_width > div->width) {
+			pr_warn("%s: bad %s fraction width for %s (%u > %u)\n",
+				__func__, field_name, clock_name,
+				div->frac_width, div->width);
+			return false;
+		}
+
+	return true;
+}
+
+/*
+ * If a clock has two dividers, the combined number of fractional
+ * bits must be representable in a 32-bit unsigned value.  This
+ * is because we scale up a dividend using both dividers before
+ * dividing to improve accuracy, and we need to avoid overflow.
+ */
+static bool kona_dividers_valid(struct kona_clk *bcm_clk)
+{
+	struct peri_clk_data *peri = bcm_clk->peri;
+	struct bcm_clk_div *div;
+	struct bcm_clk_div *pre_div;
+	u32 limit;
+
+	BUG_ON(bcm_clk->type != bcm_clk_peri);
+
+	if (!divider_exists(&peri->div) || !divider_exists(&peri->pre_div))
+		return true;
+
+	div = &peri->div;
+	pre_div = &peri->pre_div;
+	if (divider_is_fixed(div) || divider_is_fixed(pre_div))
+		return true;
+
+	limit = BITS_PER_BYTE * sizeof(u32);
+
+	return div->frac_width + pre_div->frac_width <= limit;
+}
+
+
+/* A trigger just needs to represent a valid bit position */
+static bool trig_valid(struct bcm_clk_trig *trig, const char *field_name,
+			const char *clock_name)
+{
+	return bit_posn_valid(trig->bit, field_name, clock_name);
+}
+
+/* Determine whether the set of peripheral clock registers are valid. */
+static bool
+peri_clk_data_valid(struct kona_clk *bcm_clk)
+{
+	struct peri_clk_data *peri;
+	struct bcm_clk_gate *gate;
+	struct bcm_clk_sel *sel;
+	struct bcm_clk_div *div;
+	struct bcm_clk_div *pre_div;
+	struct bcm_clk_trig *trig;
+	const char *name;
+
+	BUG_ON(bcm_clk->type != bcm_clk_peri);
+
+	/*
+	 * First validate register offsets.  This is the only place
+	 * where we need something from the ccu, so we do these
+	 * together.
+	 */
+	if (!peri_clk_data_offsets_valid(bcm_clk))
+		return false;
+
+	peri = bcm_clk->peri;
+	name = bcm_clk->name;
+	gate = &peri->gate;
+	if (gate_exists(gate) && !gate_valid(gate, "gate", name))
+		return false;
+
+	sel = &peri->sel;
+	if (selector_exists(sel)) {
+		if (!sel_valid(sel, "selector", name))
+			return false;
+
+	} else if (sel->parent_count > 1) {
+		pr_err("%s: multiple parents but no selector for %s\n",
+			__func__, name);
+
+		return false;
+	}
+
+	div = &peri->div;
+	pre_div = &peri->pre_div;
+	if (divider_exists(div)) {
+		if (!div_valid(div, "divider", name))
+			return false;
+
+		if (divider_exists(pre_div))
+			if (!div_valid(pre_div, "pre-divider", name))
+				return false;
+	} else if (divider_exists(pre_div)) {
+		pr_err("%s: pre-divider but no divider for %s\n", __func__,
+			name);
+		return false;
+	}
+
+	trig = &peri->trig;
+	if (trigger_exists(trig)) {
+		if (!trig_valid(trig, "trigger", name))
+			return false;
+
+		if (trigger_exists(&peri->pre_trig)) {
+			if (!trig_valid(trig, "pre-trigger", name)) {
+				return false;
+			}
+		}
+		if (!clk_requires_trigger(bcm_clk)) {
+			pr_warn("%s: ignoring trigger for %s (not needed)\n",
+				__func__, name);
+			trigger_clear_exists(trig);
+		}
+	} else if (trigger_exists(&peri->pre_trig)) {
+		pr_err("%s: pre-trigger but no trigger for %s\n", __func__,
+			name);
+		return false;
+	} else if (clk_requires_trigger(bcm_clk)) {
+		pr_err("%s: required trigger missing for %s\n", __func__,
+			name);
+		return false;
+	}
+
+	return kona_dividers_valid(bcm_clk);
+}
+
+static bool kona_clk_valid(struct kona_clk *bcm_clk)
+{
+	switch (bcm_clk->type) {
+	case bcm_clk_peri:
+		if (!peri_clk_data_valid(bcm_clk))
+			return false;
+		break;
+	default:
+		pr_err("%s: unrecognized clock type (%d)\n", __func__,
+			(int)bcm_clk->type);
+		return false;
+	}
+	return true;
+}
+
+/*
+ * Scan an array of parent clock names to determine whether there
+ * are any entries containing BAD_CLK_NAME.  Such entries are
+ * placeholders for non-supported clocks.  Keep track of the
+ * position of each clock name in the original array.
+ *
+ * Allocates an array of pointers to to hold the names of all
+ * non-null entries in the original array, and returns a pointer to
+ * that array in *names.  This will be used for registering the
+ * clock with the common clock code.  On successful return,
+ * *count indicates how many entries are in that names array.
+ *
+ * If there is more than one entry in the resulting names array,
+ * another array is allocated to record the parent selector value
+ * for each (defined) parent clock.  This is the value that
+ * represents this parent clock in the clock's source selector
+ * register.  The position of the clock in the original parent array
+ * defines that selector value.  The number of entries in this array
+ * is the same as the number of entries in the parent names array.
+ *
+ * The array of selector values is returned.  If the clock has no
+ * parents, no selector is required and a null pointer is returned.
+ *
+ * Returns a null pointer if the clock names array supplied was
+ * null.  (This is not an error.)
+ *
+ * Returns a pointer-coded error if an error occurs.
+ */
+static u32 *parent_process(const char *clocks[],
+			u32 *count, const char ***names)
+{
+	static const char **parent_names;
+	static u32 *parent_sel;
+	const char **clock;
+	u32 parent_count;
+	u32 bad_count = 0;
+	u32 orig_count;
+	u32 i;
+	u32 j;
+
+	*count = 0;	/* In case of early return */
+	*names = NULL;
+	if (!clocks)
+		return NULL;
+
+	/*
+	 * Count the number of names in the null-terminated array,
+	 * and find out how many of those are actually clock names.
+	 */
+	for (clock = clocks; *clock; clock++)
+		if (*clock == BAD_CLK_NAME)
+			bad_count++;
+	orig_count = (u32)(clock - clocks);
+	parent_count = orig_count - bad_count;
+
+	/* If all clocks are unsupported, we treat it as no clock */
+	if (!parent_count)
+		return NULL;
+
+	/* Avoid exceeding our parent clock limit */
+	if (parent_count > PARENT_COUNT_MAX) {
+		pr_err("%s: too many parents (%u > %u)\n", __func__,
+			parent_count, PARENT_COUNT_MAX);
+		return ERR_PTR(-EINVAL);
+	}
+
+	/*
+	 * There is one parent name for each defined parent clock.
+	 * We also maintain an array containing the selector value
+	 * for each defined clock.  If there's only one clock, the
+	 * selector is not required, but we allocate space for the
+	 * array anyway to keep things simple.
+	 */
+	parent_names = kmalloc(parent_count * sizeof(parent_names), GFP_KERNEL);
+	if (!parent_names) {
+		pr_err("%s: error allocating %u parent names\n", __func__,
+				parent_count);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* There is at least one parent, so allocate a selector array */
+
+	parent_sel = kmalloc(parent_count * sizeof(*parent_sel), GFP_KERNEL);
+	if (!parent_sel) {
+		pr_err("%s: error allocating %u parent selectors\n", __func__,
+				parent_count);
+		kfree(parent_names);
+
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* Now fill in the parent names and selector arrays */
+	for (i = 0, j = 0; i < orig_count; i++) {
+		if (clocks[i] != BAD_CLK_NAME) {
+			parent_names[j] = clocks[i];
+			parent_sel[j] = i;
+			j++;
+		}
+	}
+	*names = parent_names;
+	*count = parent_count;
+
+	return parent_sel;
+}
+
+static int
+clk_sel_setup(const char **clocks, struct bcm_clk_sel *sel,
+		struct clk_init_data *init_data)
+{
+	const char **parent_names = NULL;
+	u32 parent_count = 0;
+	u32 *parent_sel;
+
+	/*
+	 * If a peripheral clock has multiple parents, the value
+	 * used by the hardware to select that parent is represented
+	 * by the parent clock's position in the "clocks" list.  Some
+	 * values don't have defined or supported clocks; these will
+	 * have BAD_CLK_NAME entries in the parents[] array.  The
+	 * list is terminated by a NULL entry.
+	 *
+	 * We need to supply (only) the names of defined parent
+	 * clocks when registering a clock though, so we use an
+	 * array of parent selector values to map between the
+	 * indexes the common clock code uses and the selector
+	 * values we need.
+	 */
+	parent_sel = parent_process(clocks, &parent_count, &parent_names);
+	if (IS_ERR(parent_sel)) {
+		int ret = PTR_ERR(parent_sel);
+
+		pr_err("%s: error processing parent clocks for %s (%d)\n",
+			__func__, init_data->name, ret);
+
+		return ret;
+	}
+
+	init_data->parent_names = parent_names;
+	init_data->num_parents = parent_count;
+
+	sel->parent_count = parent_count;
+	sel->parent_sel = parent_sel;
+
+	return 0;
+}
+
+static void clk_sel_teardown(struct bcm_clk_sel *sel,
+		struct clk_init_data *init_data)
+{
+	kfree(sel->parent_sel);
+	sel->parent_sel = NULL;
+	sel->parent_count = 0;
+
+	init_data->num_parents = 0;
+	kfree(init_data->parent_names);
+	init_data->parent_names = NULL;
+}
+
+static void peri_clk_teardown(struct peri_clk_data *data,
+				struct clk_init_data *init_data)
+{
+	clk_sel_teardown(&data->sel, init_data);
+	init_data->ops = NULL;
+}
+
+/*
+ * Caller is responsible for freeing the parent_names[] and
+ * parent_sel[] arrays in the peripheral clock's "data" structure
+ * that can be assigned if the clock has one or more parent clocks
+ * associated with it.
+ */
+static int peri_clk_setup(struct ccu_data *ccu, struct peri_clk_data *data,
+			struct clk_init_data *init_data)
+{
+	init_data->ops = &kona_peri_clk_ops;
+	init_data->flags = CLK_IGNORE_UNUSED;
+
+	return clk_sel_setup(data->clocks, &data->sel, init_data);
+}
+
+static void bcm_clk_teardown(struct kona_clk *bcm_clk)
+{
+	switch (bcm_clk->type) {
+	case bcm_clk_peri:
+		peri_clk_teardown(bcm_clk->data, &bcm_clk->init_data);
+		break;
+	default:
+		break;
+	}
+	bcm_clk->data = NULL;
+	bcm_clk->type = bcm_clk_none;
+}
+
+static void kona_clk_teardown(struct clk *clk)
+{
+	struct clk_hw *hw;
+	struct kona_clk *bcm_clk;
+
+	if (!clk)
+		return;
+
+	hw = __clk_get_hw(clk);
+	if (!hw) {
+		pr_err("%s: clk %p has null hw pointer\n", __func__, clk);
+		return;
+	}
+	clk_unregister(clk);
+
+	bcm_clk = to_kona_clk(hw);
+	bcm_clk_teardown(bcm_clk);
+}
+
+struct clk *kona_clk_setup(struct ccu_data *ccu, const char *name,
+			enum bcm_clk_type type, void *data)
+{
+	struct kona_clk *bcm_clk;
+	struct clk_init_data *init_data;
+	struct clk *clk = NULL;
+
+	bcm_clk = kzalloc(sizeof(*bcm_clk), GFP_KERNEL);
+	if (!bcm_clk) {
+		pr_err("%s: failed to allocate bcm_clk for %s\n", __func__,
+			name);
+		return NULL;
+	}
+	bcm_clk->ccu = ccu;
+	bcm_clk->name = name;
+
+	init_data = &bcm_clk->init_data;
+	init_data->name = name;
+	switch (type) {
+	case bcm_clk_peri:
+		if (peri_clk_setup(ccu, data, init_data))
+			goto out_free;
+		break;
+	default:
+		data = NULL;
+		break;
+	}
+	bcm_clk->type = type;
+	bcm_clk->data = data;
+
+	/* Make sure everything makes sense before we set it up */
+	if (!kona_clk_valid(bcm_clk)) {
+		pr_err("%s: clock data invalid for %s\n", __func__, name);
+		goto out_teardown;
+	}
+
+	bcm_clk->hw.init = init_data;
+	clk = clk_register(NULL, &bcm_clk->hw);
+	if (IS_ERR(clk)) {
+		pr_err("%s: error registering clock %s (%ld)\n", __func__,
+				name, PTR_ERR(clk));
+		goto out_teardown;
+	}
+	BUG_ON(!clk);
+
+	return clk;
+out_teardown:
+	bcm_clk_teardown(bcm_clk);
+out_free:
+	kfree(bcm_clk);
+
+	return NULL;
+}
+
+static void ccu_clks_teardown(struct ccu_data *ccu)
+{
+	u32 i;
+
+	for (i = 0; i < ccu->data.clk_num; i++)
+		kona_clk_teardown(ccu->data.clks[i]);
+	kfree(ccu->data.clks);
+}
+
+static void kona_ccu_teardown(struct ccu_data *ccu)
+{
+	if (!ccu)
+		return;
+
+	if (!ccu->base)
+		goto done;
+
+	of_clk_del_provider(ccu->node);	/* safe if never added */
+	ccu_clks_teardown(ccu);
+	list_del(&ccu->links);
+	of_node_put(ccu->node);
+	iounmap(ccu->base);
+done:
+	kfree(ccu->name);
+	kfree(ccu);
+}
+
+/*
+ * Set up a CCU.  Call the provided ccu_clks_setup callback to
+ * initialize the array of clocks provided by the CCU.
+ */
+void __init kona_dt_ccu_setup(struct device_node *node,
+			int (*ccu_clks_setup)(struct ccu_data *))
+{
+	struct ccu_data *ccu;
+	struct resource res = { 0 };
+	resource_size_t range;
+	int ret;
+
+	ccu = kzalloc(sizeof(*ccu), GFP_KERNEL);
+	if (ccu)
+		ccu->name = kstrdup(node->name, GFP_KERNEL);
+	if (!ccu || !ccu->name) {
+		pr_err("%s: unable to allocate CCU struct for %s\n",
+			__func__, node->name);
+		kfree(ccu);
+
+		return;
+	}
+
+	ret = of_address_to_resource(node, 0, &res);
+	if (ret) {
+		pr_err("%s: no valid CCU registers found for %s\n", __func__,
+			node->name);
+		goto out_err;
+	}
+
+	range = resource_size(&res);
+	if (range > (resource_size_t)U32_MAX) {
+		pr_err("%s: address range too large for %s\n", __func__,
+			node->name);
+		goto out_err;
+	}
+
+	ccu->range = (u32)range;
+	ccu->base = ioremap(res.start, ccu->range);
+	if (!ccu->base) {
+		pr_err("%s: unable to map CCU registers for %s\n", __func__,
+			node->name);
+		goto out_err;
+	}
+
+	spin_lock_init(&ccu->lock);
+	INIT_LIST_HEAD(&ccu->links);
+	ccu->node = of_node_get(node);
+
+	list_add_tail(&ccu->links, &ccu_list);
+
+	/* Set up clocks array (in ccu->data) */
+	if (ccu_clks_setup(ccu))
+		goto out_err;
+
+	ret = of_clk_add_provider(node, of_clk_src_onecell_get, &ccu->data);
+	if (ret) {
+		pr_err("%s: error adding ccu %s as provider (%d)\n", __func__,
+				node->name, ret);
+		goto out_err;
+	}
+
+	if (!kona_ccu_init(ccu))
+		pr_err("Broadcom %s initialization had errors\n", node->name);
+
+	return;
+out_err:
+	kona_ccu_teardown(ccu);
+	pr_err("Broadcom %s setup aborted\n", node->name);
+}
diff --git a/drivers/clk/bcm/clk-kona.c b/drivers/clk/bcm/clk-kona.c
new file mode 100644
index 00000000000..e3d339e0830
--- /dev/null
+++ b/drivers/clk/bcm/clk-kona.c
@@ -0,0 +1,1033 @@
+/*
+ * Copyright (C) 2013 Broadcom Corporation
+ * Copyright 2013 Linaro Limited
+ *
+ * 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.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include "clk-kona.h"
+
+#include <linux/delay.h>
+
+#define CCU_ACCESS_PASSWORD      0xA5A500
+#define CLK_GATE_DELAY_LOOP      2000
+
+/* Bitfield operations */
+
+/* Produces a mask of set bits covering a range of a 32-bit value */
+static inline u32 bitfield_mask(u32 shift, u32 width)
+{
+	return ((1 << width) - 1) << shift;
+}
+
+/* Extract the value of a bitfield found within a given register value */
+static inline u32 bitfield_extract(u32 reg_val, u32 shift, u32 width)
+{
+	return (reg_val & bitfield_mask(shift, width)) >> shift;
+}
+
+/* Replace the value of a bitfield found within a given register value */
+static inline u32 bitfield_replace(u32 reg_val, u32 shift, u32 width, u32 val)
+{
+	u32 mask = bitfield_mask(shift, width);
+
+	return (reg_val & ~mask) | (val << shift);
+}
+
+/* Divider and scaling helpers */
+
+/*
+ * Implement DIV_ROUND_CLOSEST() for 64-bit dividend and both values
+ * unsigned.  Note that unlike do_div(), the remainder is discarded
+ * and the return value is the quotient (not the remainder).
+ */
+u64 do_div_round_closest(u64 dividend, unsigned long divisor)
+{
+	u64 result;
+
+	result = dividend + ((u64)divisor >> 1);
+	(void)do_div(result, divisor);
+
+	return result;
+}
+
+/* Convert a divider into the scaled divisor value it represents. */
+static inline u64 scaled_div_value(struct bcm_clk_div *div, u32 reg_div)
+{
+	return (u64)reg_div + ((u64)1 << div->frac_width);
+}
+
+/*
+ * Build a scaled divider value as close as possible to the
+ * given whole part (div_value) and fractional part (expressed
+ * in billionths).
+ */
+u64 scaled_div_build(struct bcm_clk_div *div, u32 div_value, u32 billionths)
+{
+	u64 combined;
+
+	BUG_ON(!div_value);
+	BUG_ON(billionths >= BILLION);
+
+	combined = (u64)div_value * BILLION + billionths;
+	combined <<= div->frac_width;
+
+	return do_div_round_closest(combined, BILLION);
+}
+
+/* The scaled minimum divisor representable by a divider */
+static inline u64
+scaled_div_min(struct bcm_clk_div *div)
+{
+	if (divider_is_fixed(div))
+		return (u64)div->fixed;
+
+	return scaled_div_value(div, 0);
+}
+
+/* The scaled maximum divisor representable by a divider */
+u64 scaled_div_max(struct bcm_clk_div *div)
+{
+	u32 reg_div;
+
+	if (divider_is_fixed(div))
+		return (u64)div->fixed;
+
+	reg_div = ((u32)1 << div->width) - 1;
+
+	return scaled_div_value(div, reg_div);
+}
+
+/*
+ * Convert a scaled divisor into its divider representation as
+ * stored in a divider register field.
+ */
+static inline u32
+divider(struct bcm_clk_div *div, u64 scaled_div)
+{
+	BUG_ON(scaled_div < scaled_div_min(div));
+	BUG_ON(scaled_div > scaled_div_max(div));
+
+	return (u32)(scaled_div - ((u64)1 << div->frac_width));
+}
+
+/* Return a rate scaled for use when dividing by a scaled divisor. */
+static inline u64
+scale_rate(struct bcm_clk_div *div, u32 rate)
+{
+	if (divider_is_fixed(div))
+		return (u64)rate;
+
+	return (u64)rate << div->frac_width;
+}
+
+/* CCU access */
+
+/* Read a 32-bit register value from a CCU's address space. */
+static inline u32 __ccu_read(struct ccu_data *ccu, u32 reg_offset)
+{
+	return readl(ccu->base + reg_offset);
+}
+
+/* Write a 32-bit register value into a CCU's address space. */
+static inline void
+__ccu_write(struct ccu_data *ccu, u32 reg_offset, u32 reg_val)
+{
+	writel(reg_val, ccu->base + reg_offset);
+}
+
+static inline unsigned long ccu_lock(struct ccu_data *ccu)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ccu->lock, flags);
+
+	return flags;
+}
+static inline void ccu_unlock(struct ccu_data *ccu, unsigned long flags)
+{
+	spin_unlock_irqrestore(&ccu->lock, flags);
+}
+
+/*
+ * Enable/disable write access to CCU protected registers.  The
+ * WR_ACCESS register for all CCUs is at offset 0.
+ */
+static inline void __ccu_write_enable(struct ccu_data *ccu)
+{
+	if (ccu->write_enabled) {
+		pr_err("%s: access already enabled for %s\n", __func__,
+			ccu->name);
+		return;
+	}
+	ccu->write_enabled = true;
+	__ccu_write(ccu, 0, CCU_ACCESS_PASSWORD | 1);
+}
+
+static inline void __ccu_write_disable(struct ccu_data *ccu)
+{
+	if (!ccu->write_enabled) {
+		pr_err("%s: access wasn't enabled for %s\n", __func__,
+			ccu->name);
+		return;
+	}
+
+	__ccu_write(ccu, 0, CCU_ACCESS_PASSWORD);
+	ccu->write_enabled = false;
+}
+
+/*
+ * Poll a register in a CCU's address space, returning when the
+ * specified bit in that register's value is set (or clear).  Delay
+ * a microsecond after each read of the register.  Returns true if
+ * successful, or false if we gave up trying.
+ *
+ * Caller must ensure the CCU lock is held.
+ */
+static inline bool
+__ccu_wait_bit(struct ccu_data *ccu, u32 reg_offset, u32 bit, bool want)
+{
+	unsigned int tries;
+	u32 bit_mask = 1 << bit;
+
+	for (tries = 0; tries < CLK_GATE_DELAY_LOOP; tries++) {
+		u32 val;
+		bool bit_val;
+
+		val = __ccu_read(ccu, reg_offset);
+		bit_val = (val & bit_mask) != 0;
+		if (bit_val == want)
+			return true;
+		udelay(1);
+	}
+	return false;
+}
+
+/* Gate operations */
+
+/* Determine whether a clock is gated.  CCU lock must be held.  */
+static bool
+__is_clk_gate_enabled(struct ccu_data *ccu, struct bcm_clk_gate *gate)
+{
+	u32 bit_mask;
+	u32 reg_val;
+
+	/* If there is no gate we can assume it's enabled. */
+	if (!gate_exists(gate))
+		return true;
+
+	bit_mask = 1 << gate->status_bit;
+	reg_val = __ccu_read(ccu, gate->offset);
+
+	return (reg_val & bit_mask) != 0;
+}
+
+/* Determine whether a clock is gated. */
+static bool
+is_clk_gate_enabled(struct ccu_data *ccu, struct bcm_clk_gate *gate)
+{
+	long flags;
+	bool ret;
+
+	/* Avoid taking the lock if we can */
+	if (!gate_exists(gate))
+		return true;
+
+	flags = ccu_lock(ccu);
+	ret = __is_clk_gate_enabled(ccu, gate);
+	ccu_unlock(ccu, flags);
+
+	return ret;
+}
+
+/*
+ * Commit our desired gate state to the hardware.
+ * Returns true if successful, false otherwise.
+ */
+static bool
+__gate_commit(struct ccu_data *ccu, struct bcm_clk_gate *gate)
+{
+	u32 reg_val;
+	u32 mask;
+	bool enabled = false;
+
+	BUG_ON(!gate_exists(gate));
+	if (!gate_is_sw_controllable(gate))
+		return true;		/* Nothing we can change */
+
+	reg_val = __ccu_read(ccu, gate->offset);
+
+	/* For a hardware/software gate, set which is in control */
+	if (gate_is_hw_controllable(gate)) {
+		mask = (u32)1 << gate->hw_sw_sel_bit;
+		if (gate_is_sw_managed(gate))
+			reg_val |= mask;
+		else
+			reg_val &= ~mask;
+	}
+
+	/*
+	 * If software is in control, enable or disable the gate.
+	 * If hardware is, clear the enabled bit for good measure.
+	 * If a software controlled gate can't be disabled, we're
+	 * required to write a 0 into the enable bit (but the gate
+	 * will be enabled).
+	 */
+	mask = (u32)1 << gate->en_bit;
+	if (gate_is_sw_managed(gate) && (enabled = gate_is_enabled(gate)) &&
+			!gate_is_no_disable(gate))
+		reg_val |= mask;
+	else
+		reg_val &= ~mask;
+
+	__ccu_write(ccu, gate->offset, reg_val);
+
+	/* For a hardware controlled gate, we're done */
+	if (!gate_is_sw_managed(gate))
+		return true;
+
+	/* Otherwise wait for the gate to be in desired state */
+	return __ccu_wait_bit(ccu, gate->offset, gate->status_bit, enabled);
+}
+
+/*
+ * Initialize a gate.  Our desired state (hardware/software select,
+ * and if software, its enable state) is committed to hardware
+ * without the usual checks to see if it's already set up that way.
+ * Returns true if successful, false otherwise.
+ */
+static bool gate_init(struct ccu_data *ccu, struct bcm_clk_gate *gate)
+{
+	if (!gate_exists(gate))
+		return true;
+	return __gate_commit(ccu, gate);
+}
+
+/*
+ * Set a gate to enabled or disabled state.  Does nothing if the
+ * gate is not currently under software control, or if it is already
+ * in the requested state.  Returns true if successful, false
+ * otherwise.  CCU lock must be held.
+ */
+static bool
+__clk_gate(struct ccu_data *ccu, struct bcm_clk_gate *gate, bool enable)
+{
+	bool ret;
+
+	if (!gate_exists(gate) || !gate_is_sw_managed(gate))
+		return true;	/* Nothing to do */
+
+	if (!enable && gate_is_no_disable(gate)) {
+		pr_warn("%s: invalid gate disable request (ignoring)\n",
+			__func__);
+		return true;
+	}
+
+	if (enable == gate_is_enabled(gate))
+		return true;	/* No change */
+
+	gate_flip_enabled(gate);
+	ret = __gate_commit(ccu, gate);
+	if (!ret)
+		gate_flip_enabled(gate);	/* Revert the change */
+
+	return ret;
+}
+
+/* Enable or disable a gate.  Returns 0 if successful, -EIO otherwise */
+static int clk_gate(struct ccu_data *ccu, const char *name,
+			struct bcm_clk_gate *gate, bool enable)
+{
+	unsigned long flags;
+	bool success;
+
+	/*
+	 * Avoid taking the lock if we can.  We quietly ignore
+	 * requests to change state that don't make sense.
+	 */
+	if (!gate_exists(gate) || !gate_is_sw_managed(gate))
+		return 0;
+	if (!enable && gate_is_no_disable(gate))
+		return 0;
+
+	flags = ccu_lock(ccu);
+	__ccu_write_enable(ccu);
+
+	success = __clk_gate(ccu, gate, enable);
+
+	__ccu_write_disable(ccu);
+	ccu_unlock(ccu, flags);
+
+	if (success)
+		return 0;
+
+	pr_err("%s: failed to %s gate for %s\n", __func__,
+		enable ? "enable" : "disable", name);
+
+	return -EIO;
+}
+
+/* Trigger operations */
+
+/*
+ * Caller must ensure CCU lock is held and access is enabled.
+ * Returns true if successful, false otherwise.
+ */
+static bool __clk_trigger(struct ccu_data *ccu, struct bcm_clk_trig *trig)
+{
+	/* Trigger the clock and wait for it to finish */
+	__ccu_write(ccu, trig->offset, 1 << trig->bit);
+
+	return __ccu_wait_bit(ccu, trig->offset, trig->bit, false);
+}
+
+/* Divider operations */
+
+/* Read a divider value and return the scaled divisor it represents. */
+static u64 divider_read_scaled(struct ccu_data *ccu, struct bcm_clk_div *div)
+{
+	unsigned long flags;
+	u32 reg_val;
+	u32 reg_div;
+
+	if (divider_is_fixed(div))
+		return (u64)div->fixed;
+
+	flags = ccu_lock(ccu);
+	reg_val = __ccu_read(ccu, div->offset);
+	ccu_unlock(ccu, flags);
+
+	/* Extract the full divider field from the register value */
+	reg_div = bitfield_extract(reg_val, div->shift, div->width);
+
+	/* Return the scaled divisor value it represents */
+	return scaled_div_value(div, reg_div);
+}
+
+/*
+ * Convert a divider's scaled divisor value into its recorded form
+ * and commit it into the hardware divider register.
+ *
+ * Returns 0 on success.  Returns -EINVAL for invalid arguments.
+ * Returns -ENXIO if gating failed, and -EIO if a trigger failed.
+ */
+static int __div_commit(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_div *div, struct bcm_clk_trig *trig)
+{
+	bool enabled;
+	u32 reg_div;
+	u32 reg_val;
+	int ret = 0;
+
+	BUG_ON(divider_is_fixed(div));
+
+	/*
+	 * If we're just initializing the divider, and no initial
+	 * state was defined in the device tree, we just find out
+	 * what its current value is rather than updating it.
+	 */
+	if (div->scaled_div == BAD_SCALED_DIV_VALUE) {
+		reg_val = __ccu_read(ccu, div->offset);
+		reg_div = bitfield_extract(reg_val, div->shift, div->width);
+		div->scaled_div = scaled_div_value(div, reg_div);
+
+		return 0;
+	}
+
+	/* Convert the scaled divisor to the value we need to record */
+	reg_div = divider(div, div->scaled_div);
+
+	/* Clock needs to be enabled before changing the rate */
+	enabled = __is_clk_gate_enabled(ccu, gate);
+	if (!enabled && !__clk_gate(ccu, gate, true)) {
+		ret = -ENXIO;
+		goto out;
+	}
+
+	/* Replace the divider value and record the result */
+	reg_val = __ccu_read(ccu, div->offset);
+	reg_val = bitfield_replace(reg_val, div->shift, div->width, reg_div);
+	__ccu_write(ccu, div->offset, reg_val);
+
+	/* If the trigger fails we still want to disable the gate */
+	if (!__clk_trigger(ccu, trig))
+		ret = -EIO;
+
+	/* Disable the clock again if it was disabled to begin with */
+	if (!enabled && !__clk_gate(ccu, gate, false))
+		ret = ret ? ret : -ENXIO;	/* return first error */
+out:
+	return ret;
+}
+
+/*
+ * Initialize a divider by committing our desired state to hardware
+ * without the usual checks to see if it's already set up that way.
+ * Returns true if successful, false otherwise.
+ */
+static bool div_init(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_div *div, struct bcm_clk_trig *trig)
+{
+	if (!divider_exists(div) || divider_is_fixed(div))
+		return true;
+	return !__div_commit(ccu, gate, div, trig);
+}
+
+static int divider_write(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_div *div, struct bcm_clk_trig *trig,
+			u64 scaled_div)
+{
+	unsigned long flags;
+	u64 previous;
+	int ret;
+
+	BUG_ON(divider_is_fixed(div));
+
+	previous = div->scaled_div;
+	if (previous == scaled_div)
+		return 0;	/* No change */
+
+	div->scaled_div = scaled_div;
+
+	flags = ccu_lock(ccu);
+	__ccu_write_enable(ccu);
+
+	ret = __div_commit(ccu, gate, div, trig);
+
+	__ccu_write_disable(ccu);
+	ccu_unlock(ccu, flags);
+
+	if (ret)
+		div->scaled_div = previous;		/* Revert the change */
+
+	return ret;
+
+}
+
+/* Common clock rate helpers */
+
+/*
+ * Implement the common clock framework recalc_rate method, taking
+ * into account a divider and an optional pre-divider.  The
+ * pre-divider register pointer may be NULL.
+ */
+static unsigned long clk_recalc_rate(struct ccu_data *ccu,
+			struct bcm_clk_div *div, struct bcm_clk_div *pre_div,
+			unsigned long parent_rate)
+{
+	u64 scaled_parent_rate;
+	u64 scaled_div;
+	u64 result;
+
+	if (!divider_exists(div))
+		return parent_rate;
+
+	if (parent_rate > (unsigned long)LONG_MAX)
+		return 0;	/* actually this would be a caller bug */
+
+	/*
+	 * If there is a pre-divider, divide the scaled parent rate
+	 * by the pre-divider value first.  In this case--to improve
+	 * accuracy--scale the parent rate by *both* the pre-divider
+	 * value and the divider before actually computing the
+	 * result of the pre-divider.
+	 *
+	 * If there's only one divider, just scale the parent rate.
+	 */
+	if (pre_div && divider_exists(pre_div)) {
+		u64 scaled_rate;
+
+		scaled_rate = scale_rate(pre_div, parent_rate);
+		scaled_rate = scale_rate(div, scaled_rate);
+		scaled_div = divider_read_scaled(ccu, pre_div);
+		scaled_parent_rate = do_div_round_closest(scaled_rate,
+							scaled_div);
+	} else  {
+		scaled_parent_rate = scale_rate(div, parent_rate);
+	}
+
+	/*
+	 * Get the scaled divisor value, and divide the scaled
+	 * parent rate by that to determine this clock's resulting
+	 * rate.
+	 */
+	scaled_div = divider_read_scaled(ccu, div);
+	result = do_div_round_closest(scaled_parent_rate, scaled_div);
+
+	return (unsigned long)result;
+}
+
+/*
+ * Compute the output rate produced when a given parent rate is fed
+ * into two dividers.  The pre-divider can be NULL, and even if it's
+ * non-null it may be nonexistent.  It's also OK for the divider to
+ * be nonexistent, and in that case the pre-divider is also ignored.
+ *
+ * If scaled_div is non-null, it is used to return the scaled divisor
+ * value used by the (downstream) divider to produce that rate.
+ */
+static long round_rate(struct ccu_data *ccu, struct bcm_clk_div *div,
+				struct bcm_clk_div *pre_div,
+				unsigned long rate, unsigned long parent_rate,
+				u64 *scaled_div)
+{
+	u64 scaled_parent_rate;
+	u64 min_scaled_div;
+	u64 max_scaled_div;
+	u64 best_scaled_div;
+	u64 result;
+
+	BUG_ON(!divider_exists(div));
+	BUG_ON(!rate);
+	BUG_ON(parent_rate > (u64)LONG_MAX);
+
+	/*
+	 * If there is a pre-divider, divide the scaled parent rate
+	 * by the pre-divider value first.  In this case--to improve
+	 * accuracy--scale the parent rate by *both* the pre-divider
+	 * value and the divider before actually computing the
+	 * result of the pre-divider.
+	 *
+	 * If there's only one divider, just scale the parent rate.
+	 *
+	 * For simplicity we treat the pre-divider as fixed (for now).
+	 */
+	if (divider_exists(pre_div)) {
+		u64 scaled_rate;
+		u64 scaled_pre_div;
+
+		scaled_rate = scale_rate(pre_div, parent_rate);
+		scaled_rate = scale_rate(div, scaled_rate);
+		scaled_pre_div = divider_read_scaled(ccu, pre_div);
+		scaled_parent_rate = do_div_round_closest(scaled_rate,
+							scaled_pre_div);
+	} else {
+		scaled_parent_rate = scale_rate(div, parent_rate);
+	}
+
+	/*
+	 * Compute the best possible divider and ensure it is in
+	 * range.  A fixed divider can't be changed, so just report
+	 * the best we can do.
+	 */
+	if (!divider_is_fixed(div)) {
+		best_scaled_div = do_div_round_closest(scaled_parent_rate,
+							rate);
+		min_scaled_div = scaled_div_min(div);
+		max_scaled_div = scaled_div_max(div);
+		if (best_scaled_div > max_scaled_div)
+			best_scaled_div = max_scaled_div;
+		else if (best_scaled_div < min_scaled_div)
+			best_scaled_div = min_scaled_div;
+	} else {
+		best_scaled_div = divider_read_scaled(ccu, div);
+	}
+
+	/* OK, figure out the resulting rate */
+	result = do_div_round_closest(scaled_parent_rate, best_scaled_div);
+
+	if (scaled_div)
+		*scaled_div = best_scaled_div;
+
+	return (long)result;
+}
+
+/* Common clock parent helpers */
+
+/*
+ * For a given parent selector (register field) value, find the
+ * index into a selector's parent_sel array that contains it.
+ * Returns the index, or BAD_CLK_INDEX if it's not found.
+ */
+static u8 parent_index(struct bcm_clk_sel *sel, u8 parent_sel)
+{
+	u8 i;
+
+	BUG_ON(sel->parent_count > (u32)U8_MAX);
+	for (i = 0; i < sel->parent_count; i++)
+		if (sel->parent_sel[i] == parent_sel)
+			return i;
+	return BAD_CLK_INDEX;
+}
+
+/*
+ * Fetch the current value of the selector, and translate that into
+ * its corresponding index in the parent array we registered with
+ * the clock framework.
+ *
+ * Returns parent array index that corresponds with the value found,
+ * or BAD_CLK_INDEX if the found value is out of range.
+ */
+static u8 selector_read_index(struct ccu_data *ccu, struct bcm_clk_sel *sel)
+{
+	unsigned long flags;
+	u32 reg_val;
+	u32 parent_sel;
+	u8 index;
+
+	/* If there's no selector, there's only one parent */
+	if (!selector_exists(sel))
+		return 0;
+
+	/* Get the value in the selector register */
+	flags = ccu_lock(ccu);
+	reg_val = __ccu_read(ccu, sel->offset);
+	ccu_unlock(ccu, flags);
+
+	parent_sel = bitfield_extract(reg_val, sel->shift, sel->width);
+
+	/* Look up that selector's parent array index and return it */
+	index = parent_index(sel, parent_sel);
+	if (index == BAD_CLK_INDEX)
+		pr_err("%s: out-of-range parent selector %u (%s 0x%04x)\n",
+			__func__, parent_sel, ccu->name, sel->offset);
+
+	return index;
+}
+
+/*
+ * Commit our desired selector value to the hardware.
+ *
+ * Returns 0 on success.  Returns -EINVAL for invalid arguments.
+ * Returns -ENXIO if gating failed, and -EIO if a trigger failed.
+ */
+static int
+__sel_commit(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_sel *sel, struct bcm_clk_trig *trig)
+{
+	u32 parent_sel;
+	u32 reg_val;
+	bool enabled;
+	int ret = 0;
+
+	BUG_ON(!selector_exists(sel));
+
+	/*
+	 * If we're just initializing the selector, and no initial
+	 * state was defined in the device tree, we just find out
+	 * what its current value is rather than updating it.
+	 */
+	if (sel->clk_index == BAD_CLK_INDEX) {
+		u8 index;
+
+		reg_val = __ccu_read(ccu, sel->offset);
+		parent_sel = bitfield_extract(reg_val, sel->shift, sel->width);
+		index = parent_index(sel, parent_sel);
+		if (index == BAD_CLK_INDEX)
+			return -EINVAL;
+		sel->clk_index = index;
+
+		return 0;
+	}
+
+	BUG_ON((u32)sel->clk_index >= sel->parent_count);
+	parent_sel = sel->parent_sel[sel->clk_index];
+
+	/* Clock needs to be enabled before changing the parent */
+	enabled = __is_clk_gate_enabled(ccu, gate);
+	if (!enabled && !__clk_gate(ccu, gate, true))
+		return -ENXIO;
+
+	/* Replace the selector value and record the result */
+	reg_val = __ccu_read(ccu, sel->offset);
+	reg_val = bitfield_replace(reg_val, sel->shift, sel->width, parent_sel);
+	__ccu_write(ccu, sel->offset, reg_val);
+
+	/* If the trigger fails we still want to disable the gate */
+	if (!__clk_trigger(ccu, trig))
+		ret = -EIO;
+
+	/* Disable the clock again if it was disabled to begin with */
+	if (!enabled && !__clk_gate(ccu, gate, false))
+		ret = ret ? ret : -ENXIO;	/* return first error */
+
+	return ret;
+}
+
+/*
+ * Initialize a selector by committing our desired state to hardware
+ * without the usual checks to see if it's already set up that way.
+ * Returns true if successful, false otherwise.
+ */
+static bool sel_init(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_sel *sel, struct bcm_clk_trig *trig)
+{
+	if (!selector_exists(sel))
+		return true;
+	return !__sel_commit(ccu, gate, sel, trig);
+}
+
+/*
+ * Write a new value into a selector register to switch to a
+ * different parent clock.  Returns 0 on success, or an error code
+ * (from __sel_commit()) otherwise.
+ */
+static int selector_write(struct ccu_data *ccu, struct bcm_clk_gate *gate,
+			struct bcm_clk_sel *sel, struct bcm_clk_trig *trig,
+			u8 index)
+{
+	unsigned long flags;
+	u8 previous;
+	int ret;
+
+	previous = sel->clk_index;
+	if (previous == index)
+		return 0;	/* No change */
+
+	sel->clk_index = index;
+
+	flags = ccu_lock(ccu);
+	__ccu_write_enable(ccu);
+
+	ret = __sel_commit(ccu, gate, sel, trig);
+
+	__ccu_write_disable(ccu);
+	ccu_unlock(ccu, flags);
+
+	if (ret)
+		sel->clk_index = previous;	/* Revert the change */
+
+	return ret;
+}
+
+/* Clock operations */
+
+static int kona_peri_clk_enable(struct clk_hw *hw)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct bcm_clk_gate *gate = &bcm_clk->peri->gate;
+
+	return clk_gate(bcm_clk->ccu, bcm_clk->name, gate, true);
+}
+
+static void kona_peri_clk_disable(struct clk_hw *hw)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct bcm_clk_gate *gate = &bcm_clk->peri->gate;
+
+	(void)clk_gate(bcm_clk->ccu, bcm_clk->name, gate, false);
+}
+
+static int kona_peri_clk_is_enabled(struct clk_hw *hw)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct bcm_clk_gate *gate = &bcm_clk->peri->gate;
+
+	return is_clk_gate_enabled(bcm_clk->ccu, gate) ? 1 : 0;
+}
+
+static unsigned long kona_peri_clk_recalc_rate(struct clk_hw *hw,
+			unsigned long parent_rate)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct peri_clk_data *data = bcm_clk->peri;
+
+	return clk_recalc_rate(bcm_clk->ccu, &data->div, &data->pre_div,
+				parent_rate);
+}
+
+static long kona_peri_clk_round_rate(struct clk_hw *hw, unsigned long rate,
+			unsigned long *parent_rate)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct bcm_clk_div *div = &bcm_clk->peri->div;
+
+	if (!divider_exists(div))
+		return __clk_get_rate(hw->clk);
+
+	/* Quietly avoid a zero rate */
+	return round_rate(bcm_clk->ccu, div, &bcm_clk->peri->pre_div,
+				rate ? rate : 1, *parent_rate, NULL);
+}
+
+static int kona_peri_clk_set_parent(struct clk_hw *hw, u8 index)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct peri_clk_data *data = bcm_clk->peri;
+	struct bcm_clk_sel *sel = &data->sel;
+	struct bcm_clk_trig *trig;
+	int ret;
+
+	BUG_ON(index >= sel->parent_count);
+
+	/* If there's only one parent we don't require a selector */
+	if (!selector_exists(sel))
+		return 0;
+
+	/*
+	 * The regular trigger is used by default, but if there's a
+	 * pre-trigger we want to use that instead.
+	 */
+	trig = trigger_exists(&data->pre_trig) ? &data->pre_trig
+					       : &data->trig;
+
+	ret = selector_write(bcm_clk->ccu, &data->gate, sel, trig, index);
+	if (ret == -ENXIO) {
+		pr_err("%s: gating failure for %s\n", __func__, bcm_clk->name);
+		ret = -EIO;	/* Don't proliferate weird errors */
+	} else if (ret == -EIO) {
+		pr_err("%s: %strigger failed for %s\n", __func__,
+			trig == &data->pre_trig ? "pre-" : "",
+			bcm_clk->name);
+	}
+
+	return ret;
+}
+
+static u8 kona_peri_clk_get_parent(struct clk_hw *hw)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct peri_clk_data *data = bcm_clk->peri;
+	u8 index;
+
+	index = selector_read_index(bcm_clk->ccu, &data->sel);
+
+	/* Not all callers would handle an out-of-range value gracefully */
+	return index == BAD_CLK_INDEX ? 0 : index;
+}
+
+static int kona_peri_clk_set_rate(struct clk_hw *hw, unsigned long rate,
+			unsigned long parent_rate)
+{
+	struct kona_clk *bcm_clk = to_kona_clk(hw);
+	struct peri_clk_data *data = bcm_clk->peri;
+	struct bcm_clk_div *div = &data->div;
+	u64 scaled_div = 0;
+	int ret;
+
+	if (parent_rate > (unsigned long)LONG_MAX)
+		return -EINVAL;
+
+	if (rate == __clk_get_rate(hw->clk))
+		return 0;
+
+	if (!divider_exists(div))
+		return rate == parent_rate ? 0 : -EINVAL;
+
+	/*
+	 * A fixed divider can't be changed.  (Nor can a fixed
+	 * pre-divider be, but for now we never actually try to
+	 * change that.)  Tolerate a request for a no-op change.
+	 */
+	if (divider_is_fixed(&data->div))
+		return rate == parent_rate ? 0 : -EINVAL;
+
+	/*
+	 * Get the scaled divisor value needed to achieve a clock
+	 * rate as close as possible to what was requested, given
+	 * the parent clock rate supplied.
+	 */
+	(void)round_rate(bcm_clk->ccu, div, &data->pre_div,
+				rate ? rate : 1, parent_rate, &scaled_div);
+
+	/*
+	 * We aren't updating any pre-divider at this point, so
+	 * we'll use the regular trigger.
+	 */
+	ret = divider_write(bcm_clk->ccu, &data->gate, &data->div,
+				&data->trig, scaled_div);
+	if (ret == -ENXIO) {
+		pr_err("%s: gating failure for %s\n", __func__, bcm_clk->name);
+		ret = -EIO;	/* Don't proliferate weird errors */
+	} else if (ret == -EIO) {
+		pr_err("%s: trigger failed for %s\n", __func__, bcm_clk->name);
+	}
+
+	return ret;
+}
+
+struct clk_ops kona_peri_clk_ops = {
+	.enable = kona_peri_clk_enable,
+	.disable = kona_peri_clk_disable,
+	.is_enabled = kona_peri_clk_is_enabled,
+	.recalc_rate = kona_peri_clk_recalc_rate,
+	.round_rate = kona_peri_clk_round_rate,
+	.set_parent = kona_peri_clk_set_parent,
+	.get_parent = kona_peri_clk_get_parent,
+	.set_rate = kona_peri_clk_set_rate,
+};
+
+/* Put a peripheral clock into its initial state */
+static bool __peri_clk_init(struct kona_clk *bcm_clk)
+{
+	struct ccu_data *ccu = bcm_clk->ccu;
+	struct peri_clk_data *peri = bcm_clk->peri;
+	const char *name = bcm_clk->name;
+	struct bcm_clk_trig *trig;
+
+	BUG_ON(bcm_clk->type != bcm_clk_peri);
+
+	if (!gate_init(ccu, &peri->gate)) {
+		pr_err("%s: error initializing gate for %s\n", __func__, name);
+		return false;
+	}
+	if (!div_init(ccu, &peri->gate, &peri->div, &peri->trig)) {
+		pr_err("%s: error initializing divider for %s\n", __func__,
+			name);
+		return false;
+	}
+
+	/*
+	 * For the pre-divider and selector, the pre-trigger is used
+	 * if it's present, otherwise we just use the regular trigger.
+	 */
+	trig = trigger_exists(&peri->pre_trig) ? &peri->pre_trig
+					       : &peri->trig;
+
+	if (!div_init(ccu, &peri->gate, &peri->pre_div, trig)) {
+		pr_err("%s: error initializing pre-divider for %s\n", __func__,
+			name);
+		return false;
+	}
+
+	if (!sel_init(ccu, &peri->gate, &peri->sel, trig)) {
+		pr_err("%s: error initializing selector for %s\n", __func__,
+			name);
+		return false;
+	}
+
+	return true;
+}
+
+static bool __kona_clk_init(struct kona_clk *bcm_clk)
+{
+	switch (bcm_clk->type) {
+	case bcm_clk_peri:
+		return __peri_clk_init(bcm_clk);
+	default:
+		BUG();
+	}
+	return -EINVAL;
+}
+
+/* Set a CCU and all its clocks into their desired initial state */
+bool __init kona_ccu_init(struct ccu_data *ccu)
+{
+	unsigned long flags;
+	unsigned int which;
+	struct clk **clks = ccu->data.clks;
+	bool success = true;
+
+	flags = ccu_lock(ccu);
+	__ccu_write_enable(ccu);
+
+	for (which = 0; which < ccu->data.clk_num; which++) {
+		struct kona_clk *bcm_clk;
+
+		if (!clks[which])
+			continue;
+		bcm_clk = to_kona_clk(__clk_get_hw(clks[which]));
+		success &= __kona_clk_init(bcm_clk);
+	}
+
+	__ccu_write_disable(ccu);
+	ccu_unlock(ccu, flags);
+	return success;
+}
diff --git a/drivers/clk/bcm/clk-kona.h b/drivers/clk/bcm/clk-kona.h
new file mode 100644
index 00000000000..5e139adc3dc
--- /dev/null
+++ b/drivers/clk/bcm/clk-kona.h
@@ -0,0 +1,410 @@
+/*
+ * Copyright (C) 2013 Broadcom Corporation
+ * Copyright 2013 Linaro Limited
+ *
+ * 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.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef _CLK_KONA_H
+#define _CLK_KONA_H
+
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/device.h>
+#include <linux/of.h>
+#include <linux/clk-provider.h>
+
+#define	BILLION		1000000000
+
+/* The common clock framework uses u8 to represent a parent index */
+#define PARENT_COUNT_MAX	((u32)U8_MAX)
+
+#define BAD_CLK_INDEX		U8_MAX	/* Can't ever be valid */
+#define BAD_CLK_NAME		((const char *)-1)
+
+#define BAD_SCALED_DIV_VALUE	U64_MAX
+
+/*
+ * Utility macros for object flag management.  If possible, flags
+ * should be defined such that 0 is the desired default value.
+ */
+#define FLAG(type, flag)		BCM_CLK_ ## type ## _FLAGS_ ## flag
+#define FLAG_SET(obj, type, flag)	((obj)->flags |= FLAG(type, flag))
+#define FLAG_CLEAR(obj, type, flag)	((obj)->flags &= ~(FLAG(type, flag)))
+#define FLAG_FLIP(obj, type, flag)	((obj)->flags ^= FLAG(type, flag))
+#define FLAG_TEST(obj, type, flag)	(!!((obj)->flags & FLAG(type, flag)))
+
+/* Clock field state tests */
+
+#define gate_exists(gate)		FLAG_TEST(gate, GATE, EXISTS)
+#define gate_is_enabled(gate)		FLAG_TEST(gate, GATE, ENABLED)
+#define gate_is_hw_controllable(gate)	FLAG_TEST(gate, GATE, HW)
+#define gate_is_sw_controllable(gate)	FLAG_TEST(gate, GATE, SW)
+#define gate_is_sw_managed(gate)	FLAG_TEST(gate, GATE, SW_MANAGED)
+#define gate_is_no_disable(gate)	FLAG_TEST(gate, GATE, NO_DISABLE)
+
+#define gate_flip_enabled(gate)		FLAG_FLIP(gate, GATE, ENABLED)
+
+#define divider_exists(div)		FLAG_TEST(div, DIV, EXISTS)
+#define divider_is_fixed(div)		FLAG_TEST(div, DIV, FIXED)
+#define divider_has_fraction(div)	(!divider_is_fixed(div) && \
+						(div)->frac_width > 0)
+
+#define selector_exists(sel)		((sel)->width != 0)
+#define trigger_exists(trig)		FLAG_TEST(trig, TRIG, EXISTS)
+
+/* Clock type, used to tell common block what it's part of */
+enum bcm_clk_type {
+	bcm_clk_none,		/* undefined clock type */
+	bcm_clk_bus,
+	bcm_clk_core,
+	bcm_clk_peri
+};
+
+/*
+ * Each CCU defines a mapped area of memory containing registers
+ * used to manage clocks implemented by the CCU.  Access to memory
+ * within the CCU's space is serialized by a spinlock.  Before any
+ * (other) address can be written, a special access "password" value
+ * must be written to its WR_ACCESS register (located at the base
+ * address of the range).  We keep track of the name of each CCU as
+ * it is set up, and maintain them in a list.
+ */
+struct ccu_data {
+	void __iomem *base;	/* base of mapped address space */
+	spinlock_t lock;	/* serialization lock */
+	bool write_enabled;	/* write access is currently enabled */
+	struct list_head links;	/* for ccu_list */
+	struct device_node *node;
+	struct clk_onecell_data data;
+	const char *name;
+	u32 range;		/* byte range of address space */
+};
+
+/*
+ * Gating control and status is managed by a 32-bit gate register.
+ *
+ * There are several types of gating available:
+ * - (no gate)
+ *     A clock with no gate is assumed to be always enabled.
+ * - hardware-only gating (auto-gating)
+ *     Enabling or disabling clocks with this type of gate is
+ *     managed automatically by the hardware.  Such clocks can be
+ *     considered by the software to be enabled.  The current status
+ *     of auto-gated clocks can be read from the gate status bit.
+ * - software-only gating
+ *     Auto-gating is not available for this type of clock.
+ *     Instead, software manages whether it's enabled by setting or
+ *     clearing the enable bit.  The current gate status of a gate
+ *     under software control can be read from the gate status bit.
+ *     To ensure a change to the gating status is complete, the
+ *     status bit can be polled to verify that the gate has entered
+ *     the desired state.
+ * - selectable hardware or software gating
+ *     Gating for this type of clock can be configured to be either
+ *     under software or hardware control.  Which type is in use is
+ *     determined by the hw_sw_sel bit of the gate register.
+ */
+struct bcm_clk_gate {
+	u32 offset;		/* gate register offset */
+	u32 status_bit;		/* 0: gate is disabled; 0: gatge is enabled */
+	u32 en_bit;		/* 0: disable; 1: enable */
+	u32 hw_sw_sel_bit;	/* 0: hardware gating; 1: software gating */
+	u32 flags;		/* BCM_CLK_GATE_FLAGS_* below */
+};
+
+/*
+ * Gate flags:
+ *   HW         means this gate can be auto-gated
+ *   SW         means the state of this gate can be software controlled
+ *   NO_DISABLE means this gate is (only) enabled if under software control
+ *   SW_MANAGED means the status of this gate is under software control
+ *   ENABLED    means this software-managed gate is *supposed* to be enabled
+ */
+#define BCM_CLK_GATE_FLAGS_EXISTS	((u32)1 << 0)	/* Gate is valid */
+#define BCM_CLK_GATE_FLAGS_HW		((u32)1 << 1)	/* Can auto-gate */
+#define BCM_CLK_GATE_FLAGS_SW		((u32)1 << 2)	/* Software control */
+#define BCM_CLK_GATE_FLAGS_NO_DISABLE	((u32)1 << 3)	/* HW or enabled */
+#define BCM_CLK_GATE_FLAGS_SW_MANAGED	((u32)1 << 4)	/* SW now in control */
+#define BCM_CLK_GATE_FLAGS_ENABLED	((u32)1 << 5)	/* If SW_MANAGED */
+
+/*
+ * Gate initialization macros.
+ *
+ * Any gate initially under software control will be enabled.
+ */
+
+/* A hardware/software gate initially under software control */
+#define HW_SW_GATE(_offset, _status_bit, _en_bit, _hw_sw_sel_bit)	\
+	{								\
+		.offset = (_offset),					\
+		.status_bit = (_status_bit),				\
+		.en_bit = (_en_bit),					\
+		.hw_sw_sel_bit = (_hw_sw_sel_bit),			\
+		.flags = FLAG(GATE, HW)|FLAG(GATE, SW)|			\
+			FLAG(GATE, SW_MANAGED)|FLAG(GATE, ENABLED)|	\
+			FLAG(GATE, EXISTS),				\
+	}
+
+/* A hardware/software gate initially under hardware control */
+#define HW_SW_GATE_AUTO(_offset, _status_bit, _en_bit, _hw_sw_sel_bit)	\
+	{								\
+		.offset = (_offset),					\
+		.status_bit = (_status_bit),				\
+		.en_bit = (_en_bit),					\
+		.hw_sw_sel_bit = (_hw_sw_sel_bit),			\
+		.flags = FLAG(GATE, HW)|FLAG(GATE, SW)|			\
+			FLAG(GATE, EXISTS),				\
+	}
+
+/* A hardware-or-enabled gate (enabled if not under hardware control) */
+#define HW_ENABLE_GATE(_offset, _status_bit, _en_bit, _hw_sw_sel_bit)	\
+	{								\
+		.offset = (_offset),					\
+		.status_bit = (_status_bit),				\
+		.en_bit = (_en_bit),					\
+		.hw_sw_sel_bit = (_hw_sw_sel_bit),			\
+		.flags = FLAG(GATE, HW)|FLAG(GATE, SW)|			\
+			FLAG(GATE, NO_DISABLE)|FLAG(GATE, EXISTS),	\
+	}
+
+/* A software-only gate */
+#define SW_ONLY_GATE(_offset, _status_bit, _en_bit)			\
+	{								\
+		.offset = (_offset),					\
+		.status_bit = (_status_bit),				\
+		.en_bit = (_en_bit),					\
+		.flags = FLAG(GATE, SW)|FLAG(GATE, SW_MANAGED)|		\
+			FLAG(GATE, ENABLED)|FLAG(GATE, EXISTS),		\
+	}
+
+/* A hardware-only gate */
+#define HW_ONLY_GATE(_offset, _status_bit)				\
+	{								\
+		.offset = (_offset),					\
+		.status_bit = (_status_bit),				\
+		.flags = FLAG(GATE, HW)|FLAG(GATE, EXISTS),		\
+	}
+
+/*
+ * Each clock can have zero, one, or two dividers which change the
+ * output rate of the clock.  Each divider can be either fixed or
+ * variable.  If there are two dividers, they are the "pre-divider"
+ * and the "regular" or "downstream" divider.  If there is only one,
+ * there is no pre-divider.
+ *
+ * A fixed divider is any non-zero (positive) value, and it
+ * indicates how the input rate is affected by the divider.
+ *
+ * The value of a variable divider is maintained in a sub-field of a
+ * 32-bit divider register.  The position of the field in the
+ * register is defined by its offset and width.  The value recorded
+ * in this field is always 1 less than the value it represents.
+ *
+ * In addition, a variable divider can indicate that some subset
+ * of its bits represent a "fractional" part of the divider.  Such
+ * bits comprise the low-order portion of the divider field, and can
+ * be viewed as representing the portion of the divider that lies to
+ * the right of the decimal point.  Most variable dividers have zero
+ * fractional bits.  Variable dividers with non-zero fraction width
+ * still record a value 1 less than the value they represent; the
+ * added 1 does *not* affect the low-order bit in this case, it
+ * affects the bits above the fractional part only.  (Often in this
+ * code a divider field value is distinguished from the value it
+ * represents by referring to the latter as a "divisor".)
+ *
+ * In order to avoid dealing with fractions, divider arithmetic is
+ * performed using "scaled" values.  A scaled value is one that's
+ * been left-shifted by the fractional width of a divider.  Dividing
+ * a scaled value by a scaled divisor produces the desired quotient
+ * without loss of precision and without any other special handling
+ * for fractions.
+ *
+ * The recorded value of a variable divider can be modified.  To
+ * modify either divider (or both), a clock must be enabled (i.e.,
+ * using its gate).  In addition, a trigger register (described
+ * below) must be used to commit the change, and polled to verify
+ * the change is complete.
+ */
+struct bcm_clk_div {
+	union {
+		struct {	/* variable divider */
+			u32 offset;	/* divider register offset */
+			u32 shift;	/* field shift */
+			u32 width;	/* field width */
+			u32 frac_width;	/* field fraction width */
+
+			u64 scaled_div;	/* scaled divider value */
+		};
+		u32 fixed;	/* non-zero fixed divider value */
+	};
+	u32 flags;		/* BCM_CLK_DIV_FLAGS_* below */
+};
+
+/*
+ * Divider flags:
+ *   EXISTS means this divider exists
+ *   FIXED means it is a fixed-rate divider
+ */
+#define BCM_CLK_DIV_FLAGS_EXISTS	((u32)1 << 0)	/* Divider is valid */
+#define BCM_CLK_DIV_FLAGS_FIXED		((u32)1 << 1)	/* Fixed-value */
+
+/* Divider initialization macros */
+
+/* A fixed (non-zero) divider */
+#define FIXED_DIVIDER(_value)						\
+	{								\
+		.fixed = (_value),					\
+		.flags = FLAG(DIV, EXISTS)|FLAG(DIV, FIXED),		\
+	}
+
+/* A divider with an integral divisor */
+#define DIVIDER(_offset, _shift, _width)				\
+	{								\
+		.offset = (_offset),					\
+		.shift = (_shift),					\
+		.width = (_width),					\
+		.scaled_div = BAD_SCALED_DIV_VALUE,			\
+		.flags = FLAG(DIV, EXISTS),				\
+	}
+
+/* A divider whose divisor has an integer and fractional part */
+#define FRAC_DIVIDER(_offset, _shift, _width, _frac_width)		\
+	{								\
+		.offset = (_offset),					\
+		.shift = (_shift),					\
+		.width = (_width),					\
+		.frac_width = (_frac_width),				\
+		.scaled_div = BAD_SCALED_DIV_VALUE,			\
+		.flags = FLAG(DIV, EXISTS),				\
+	}
+
+/*
+ * Clocks may have multiple "parent" clocks.  If there is more than
+ * one, a selector must be specified to define which of the parent
+ * clocks is currently in use.  The selected clock is indicated in a
+ * sub-field of a 32-bit selector register.  The range of
+ * representable selector values typically exceeds the number of
+ * available parent clocks.  Occasionally the reset value of a
+ * selector field is explicitly set to a (specific) value that does
+ * not correspond to a defined input clock.
+ *
+ * We register all known parent clocks with the common clock code
+ * using a packed array (i.e., no empty slots) of (parent) clock
+ * names, and refer to them later using indexes into that array.
+ * We maintain an array of selector values indexed by common clock
+ * index values in order to map between these common clock indexes
+ * and the selector values used by the hardware.
+ *
+ * Like dividers, a selector can be modified, but to do so a clock
+ * must be enabled, and a trigger must be used to commit the change.
+ */
+struct bcm_clk_sel {
+	u32 offset;		/* selector register offset */
+	u32 shift;		/* field shift */
+	u32 width;		/* field width */
+
+	u32 parent_count;	/* number of entries in parent_sel[] */
+	u32 *parent_sel;	/* array of parent selector values */
+	u8 clk_index;		/* current selected index in parent_sel[] */
+};
+
+/* Selector initialization macro */
+#define SELECTOR(_offset, _shift, _width)				\
+	{								\
+		.offset = (_offset),					\
+		.shift = (_shift),					\
+		.width = (_width),					\
+		.clk_index = BAD_CLK_INDEX,				\
+	}
+
+/*
+ * Making changes to a variable divider or a selector for a clock
+ * requires the use of a trigger.  A trigger is defined by a single
+ * bit within a register.  To signal a change, a 1 is written into
+ * that bit.  To determine when the change has been completed, that
+ * trigger bit is polled; the read value will be 1 while the change
+ * is in progress, and 0 when it is complete.
+ *
+ * Occasionally a clock will have more than one trigger.  In this
+ * case, the "pre-trigger" will be used when changing a clock's
+ * selector and/or its pre-divider.
+ */
+struct bcm_clk_trig {
+	u32 offset;		/* trigger register offset */
+	u32 bit;		/* trigger bit */
+	u32 flags;		/* BCM_CLK_TRIG_FLAGS_* below */
+};
+
+/*
+ * Trigger flags:
+ *   EXISTS means this trigger exists
+ */
+#define BCM_CLK_TRIG_FLAGS_EXISTS	((u32)1 << 0)	/* Trigger is valid */
+
+/* Trigger initialization macro */
+#define TRIGGER(_offset, _bit)						\
+	{								\
+		.offset = (_offset),					\
+		.bit = (_bit),						\
+		.flags = FLAG(TRIG, EXISTS),				\
+	}
+
+struct peri_clk_data {
+	struct bcm_clk_gate gate;
+	struct bcm_clk_trig pre_trig;
+	struct bcm_clk_div pre_div;
+	struct bcm_clk_trig trig;
+	struct bcm_clk_div div;
+	struct bcm_clk_sel sel;
+	const char *clocks[];	/* must be last; use CLOCKS() to declare */
+};
+#define CLOCKS(...)	{ __VA_ARGS__, NULL, }
+#define NO_CLOCKS	{ NULL, }	/* Must use of no parent clocks */
+
+struct kona_clk {
+	struct clk_hw hw;
+	struct clk_init_data init_data;
+	const char *name;	/* name of this clock */
+	struct ccu_data *ccu;	/* ccu this clock is associated with */
+	enum bcm_clk_type type;
+	union {
+		void *data;
+		struct peri_clk_data *peri;
+	};
+};
+#define to_kona_clk(_hw) \
+	container_of(_hw, struct kona_clk, hw)
+
+/* Exported globals */
+
+extern struct clk_ops kona_peri_clk_ops;
+
+/* Help functions */
+
+#define PERI_CLK_SETUP(clks, ccu, id, name) \
+	clks[id] = kona_clk_setup(ccu, #name, bcm_clk_peri, &name ## _data)
+
+/* Externally visible functions */
+
+extern u64 do_div_round_closest(u64 dividend, unsigned long divisor);
+extern u64 scaled_div_max(struct bcm_clk_div *div);
+extern u64 scaled_div_build(struct bcm_clk_div *div, u32 div_value,
+				u32 billionths);
+
+extern struct clk *kona_clk_setup(struct ccu_data *ccu, const char *name,
+			enum bcm_clk_type type, void *data);
+extern void __init kona_dt_ccu_setup(struct device_node *node,
+			int (*ccu_clks_setup)(struct ccu_data *));
+extern bool __init kona_ccu_init(struct ccu_data *ccu);
+
+#endif /* _CLK_KONA_H */