1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Marvell Armada 37xx SoC Peripheral clocks
4  *
5  * Copyright (C) 2016 Marvell
6  *
7  * Gregory CLEMENT <gregory.clement@free-electrons.com>
8  *
9  * Most of the peripheral clocks can be modelled like this:
10  *             _____    _______    _______
11  * TBG-A-P  --|     |  |       |  |       |   ______
12  * TBG-B-P  --| Mux |--| /div1 |--| /div2 |--| Gate |--> perip_clk
13  * TBG-A-S  --|     |  |       |  |       |  |______|
14  * TBG-B-S  --|_____|  |_______|  |_______|
15  *
16  * However some clocks may use only one or two block or and use the
17  * xtal clock as parent.
18  */
19 
20 #include <linux/clk-provider.h>
21 #include <linux/mfd/syscon.h>
22 #include <linux/of.h>
23 #include <linux/of_device.h>
24 #include <linux/platform_device.h>
25 #include <linux/regmap.h>
26 #include <linux/slab.h>
27 
28 #define TBG_SEL		0x0
29 #define DIV_SEL0	0x4
30 #define DIV_SEL1	0x8
31 #define DIV_SEL2	0xC
32 #define CLK_SEL		0x10
33 #define CLK_DIS		0x14
34 
35 #define  ARMADA_37XX_DVFS_LOAD_1 1
36 #define LOAD_LEVEL_NR	4
37 
38 #define ARMADA_37XX_NB_L0L1	0x18
39 #define ARMADA_37XX_NB_L2L3	0x1C
40 #define		ARMADA_37XX_NB_TBG_DIV_OFF	13
41 #define		ARMADA_37XX_NB_TBG_DIV_MASK	0x7
42 #define		ARMADA_37XX_NB_CLK_SEL_OFF	11
43 #define		ARMADA_37XX_NB_CLK_SEL_MASK	0x1
44 #define		ARMADA_37XX_NB_TBG_SEL_OFF	9
45 #define		ARMADA_37XX_NB_TBG_SEL_MASK	0x3
46 #define		ARMADA_37XX_NB_CONFIG_SHIFT	16
47 #define ARMADA_37XX_NB_DYN_MOD	0x24
48 #define		ARMADA_37XX_NB_DFS_EN	31
49 #define ARMADA_37XX_NB_CPU_LOAD	0x30
50 #define		ARMADA_37XX_NB_CPU_LOAD_MASK	0x3
51 #define		ARMADA_37XX_DVFS_LOAD_0		0
52 #define		ARMADA_37XX_DVFS_LOAD_1		1
53 #define		ARMADA_37XX_DVFS_LOAD_2		2
54 #define		ARMADA_37XX_DVFS_LOAD_3		3
55 
56 struct clk_periph_driver_data {
57 	struct clk_hw_onecell_data *hw_data;
58 	spinlock_t lock;
59 	void __iomem *reg;
60 
61 	/* Storage registers for suspend/resume operations */
62 	u32 tbg_sel;
63 	u32 div_sel0;
64 	u32 div_sel1;
65 	u32 div_sel2;
66 	u32 clk_sel;
67 	u32 clk_dis;
68 };
69 
70 struct clk_double_div {
71 	struct clk_hw hw;
72 	void __iomem *reg1;
73 	u8 shift1;
74 	void __iomem *reg2;
75 	u8 shift2;
76 };
77 
78 struct clk_pm_cpu {
79 	struct clk_hw hw;
80 	void __iomem *reg_mux;
81 	u8 shift_mux;
82 	u32 mask_mux;
83 	void __iomem *reg_div;
84 	u8 shift_div;
85 	struct regmap *nb_pm_base;
86 };
87 
88 #define to_clk_double_div(_hw) container_of(_hw, struct clk_double_div, hw)
89 #define to_clk_pm_cpu(_hw) container_of(_hw, struct clk_pm_cpu, hw)
90 
91 struct clk_periph_data {
92 	const char *name;
93 	const char * const *parent_names;
94 	int num_parents;
95 	struct clk_hw *mux_hw;
96 	struct clk_hw *rate_hw;
97 	struct clk_hw *gate_hw;
98 	struct clk_hw *muxrate_hw;
99 	bool is_double_div;
100 };
101 
102 static const struct clk_div_table clk_table6[] = {
103 	{ .val = 1, .div = 1, },
104 	{ .val = 2, .div = 2, },
105 	{ .val = 3, .div = 3, },
106 	{ .val = 4, .div = 4, },
107 	{ .val = 5, .div = 5, },
108 	{ .val = 6, .div = 6, },
109 	{ .val = 0, .div = 0, }, /* last entry */
110 };
111 
112 static const struct clk_div_table clk_table1[] = {
113 	{ .val = 0, .div = 1, },
114 	{ .val = 1, .div = 2, },
115 	{ .val = 0, .div = 0, }, /* last entry */
116 };
117 
118 static const struct clk_div_table clk_table2[] = {
119 	{ .val = 0, .div = 2, },
120 	{ .val = 1, .div = 4, },
121 	{ .val = 0, .div = 0, }, /* last entry */
122 };
123 
124 static const struct clk_ops clk_double_div_ops;
125 static const struct clk_ops clk_pm_cpu_ops;
126 
127 #define PERIPH_GATE(_name, _bit)		\
128 struct clk_gate gate_##_name = {		\
129 	.reg = (void *)CLK_DIS,			\
130 	.bit_idx = _bit,			\
131 	.hw.init = &(struct clk_init_data){	\
132 		.ops =  &clk_gate_ops,		\
133 	}					\
134 };
135 
136 #define PERIPH_MUX(_name, _shift)		\
137 struct clk_mux mux_##_name = {			\
138 	.reg = (void *)TBG_SEL,			\
139 	.shift = _shift,			\
140 	.mask = 3,				\
141 	.hw.init = &(struct clk_init_data){	\
142 		.ops =  &clk_mux_ro_ops,	\
143 	}					\
144 };
145 
146 #define PERIPH_DOUBLEDIV(_name, _reg1, _reg2, _shift1, _shift2)	\
147 struct clk_double_div rate_##_name = {		\
148 	.reg1 = (void *)_reg1,			\
149 	.reg2 = (void *)_reg2,			\
150 	.shift1 = _shift1,			\
151 	.shift2 = _shift2,			\
152 	.hw.init = &(struct clk_init_data){	\
153 		.ops =  &clk_double_div_ops,	\
154 	}					\
155 };
156 
157 #define PERIPH_DIV(_name, _reg, _shift, _table)	\
158 struct clk_divider rate_##_name = {		\
159 	.reg = (void *)_reg,			\
160 	.table = _table,			\
161 	.shift = _shift,			\
162 	.hw.init = &(struct clk_init_data){	\
163 		.ops =  &clk_divider_ro_ops,	\
164 	}					\
165 };
166 
167 #define PERIPH_PM_CPU(_name, _shift1, _reg, _shift2)	\
168 struct clk_pm_cpu muxrate_##_name = {		\
169 	.reg_mux = (void *)TBG_SEL,		\
170 	.mask_mux = 3,				\
171 	.shift_mux = _shift1,			\
172 	.reg_div = (void *)_reg,		\
173 	.shift_div = _shift2,			\
174 	.hw.init = &(struct clk_init_data){	\
175 		.ops =  &clk_pm_cpu_ops,	\
176 	}					\
177 };
178 
179 #define PERIPH_CLK_FULL_DD(_name, _bit, _shift, _reg1, _reg2, _shift1, _shift2)\
180 static PERIPH_GATE(_name, _bit);			    \
181 static PERIPH_MUX(_name, _shift);			    \
182 static PERIPH_DOUBLEDIV(_name, _reg1, _reg2, _shift1, _shift2);
183 
184 #define PERIPH_CLK_FULL(_name, _bit, _shift, _reg, _shift1, _table)	\
185 static PERIPH_GATE(_name, _bit);			    \
186 static PERIPH_MUX(_name, _shift);			    \
187 static PERIPH_DIV(_name, _reg, _shift1, _table);
188 
189 #define PERIPH_CLK_GATE_DIV(_name, _bit,  _reg, _shift, _table)	\
190 static PERIPH_GATE(_name, _bit);			\
191 static PERIPH_DIV(_name, _reg, _shift, _table);
192 
193 #define PERIPH_CLK_MUX_DD(_name, _shift, _reg1, _reg2, _shift1, _shift2)\
194 static PERIPH_MUX(_name, _shift);			    \
195 static PERIPH_DOUBLEDIV(_name, _reg1, _reg2, _shift1, _shift2);
196 
197 #define REF_CLK_FULL(_name)				\
198 	{ .name = #_name,				\
199 	  .parent_names = (const char *[]){ "TBG-A-P",	\
200 	      "TBG-B-P", "TBG-A-S", "TBG-B-S"},		\
201 	  .num_parents = 4,				\
202 	  .mux_hw = &mux_##_name.hw,			\
203 	  .gate_hw = &gate_##_name.hw,			\
204 	  .rate_hw = &rate_##_name.hw,			\
205 	}
206 
207 #define REF_CLK_FULL_DD(_name)				\
208 	{ .name = #_name,				\
209 	  .parent_names = (const char *[]){ "TBG-A-P",	\
210 	      "TBG-B-P", "TBG-A-S", "TBG-B-S"},		\
211 	  .num_parents = 4,				\
212 	  .mux_hw = &mux_##_name.hw,			\
213 	  .gate_hw = &gate_##_name.hw,			\
214 	  .rate_hw = &rate_##_name.hw,			\
215 	  .is_double_div = true,			\
216 	}
217 
218 #define REF_CLK_GATE(_name, _parent_name)			\
219 	{ .name = #_name,					\
220 	  .parent_names = (const char *[]){ _parent_name},	\
221 	  .num_parents = 1,					\
222 	  .gate_hw = &gate_##_name.hw,				\
223 	}
224 
225 #define REF_CLK_GATE_DIV(_name, _parent_name)			\
226 	{ .name = #_name,					\
227 	  .parent_names = (const char *[]){ _parent_name},	\
228 	  .num_parents = 1,					\
229 	  .gate_hw = &gate_##_name.hw,				\
230 	  .rate_hw = &rate_##_name.hw,				\
231 	}
232 
233 #define REF_CLK_PM_CPU(_name)				\
234 	{ .name = #_name,				\
235 	  .parent_names = (const char *[]){ "TBG-A-P",	\
236 	      "TBG-B-P", "TBG-A-S", "TBG-B-S"},		\
237 	  .num_parents = 4,				\
238 	  .muxrate_hw = &muxrate_##_name.hw,		\
239 	}
240 
241 #define REF_CLK_MUX_DD(_name)				\
242 	{ .name = #_name,				\
243 	  .parent_names = (const char *[]){ "TBG-A-P",	\
244 	      "TBG-B-P", "TBG-A-S", "TBG-B-S"},		\
245 	  .num_parents = 4,				\
246 	  .mux_hw = &mux_##_name.hw,			\
247 	  .rate_hw = &rate_##_name.hw,			\
248 	  .is_double_div = true,			\
249 	}
250 
251 /* NB periph clocks */
252 PERIPH_CLK_FULL_DD(mmc, 2, 0, DIV_SEL2, DIV_SEL2, 16, 13);
253 PERIPH_CLK_FULL_DD(sata_host, 3, 2, DIV_SEL2, DIV_SEL2, 10, 7);
254 PERIPH_CLK_FULL_DD(sec_at, 6, 4, DIV_SEL1, DIV_SEL1, 3, 0);
255 PERIPH_CLK_FULL_DD(sec_dap, 7, 6, DIV_SEL1, DIV_SEL1, 9, 6);
256 PERIPH_CLK_FULL_DD(tscem, 8, 8, DIV_SEL1, DIV_SEL1, 15, 12);
257 PERIPH_CLK_FULL(tscem_tmx, 10, 10, DIV_SEL1, 18, clk_table6);
258 static PERIPH_GATE(avs, 11);
259 PERIPH_CLK_FULL_DD(pwm, 13, 14, DIV_SEL0, DIV_SEL0, 3, 0);
260 PERIPH_CLK_FULL_DD(sqf, 12, 12, DIV_SEL1, DIV_SEL1, 27, 24);
261 static PERIPH_GATE(i2c_2, 16);
262 static PERIPH_GATE(i2c_1, 17);
263 PERIPH_CLK_GATE_DIV(ddr_phy, 19, DIV_SEL0, 18, clk_table2);
264 PERIPH_CLK_FULL_DD(ddr_fclk, 21, 16, DIV_SEL0, DIV_SEL0, 15, 12);
265 PERIPH_CLK_FULL(trace, 22, 18, DIV_SEL0, 20, clk_table6);
266 PERIPH_CLK_FULL(counter, 23, 20, DIV_SEL0, 23, clk_table6);
267 PERIPH_CLK_FULL_DD(eip97, 24, 24, DIV_SEL2, DIV_SEL2, 22, 19);
268 static PERIPH_PM_CPU(cpu, 22, DIV_SEL0, 28);
269 
270 static struct clk_periph_data data_nb[] = {
271 	REF_CLK_FULL_DD(mmc),
272 	REF_CLK_FULL_DD(sata_host),
273 	REF_CLK_FULL_DD(sec_at),
274 	REF_CLK_FULL_DD(sec_dap),
275 	REF_CLK_FULL_DD(tscem),
276 	REF_CLK_FULL(tscem_tmx),
277 	REF_CLK_GATE(avs, "xtal"),
278 	REF_CLK_FULL_DD(sqf),
279 	REF_CLK_FULL_DD(pwm),
280 	REF_CLK_GATE(i2c_2, "xtal"),
281 	REF_CLK_GATE(i2c_1, "xtal"),
282 	REF_CLK_GATE_DIV(ddr_phy, "TBG-A-S"),
283 	REF_CLK_FULL_DD(ddr_fclk),
284 	REF_CLK_FULL(trace),
285 	REF_CLK_FULL(counter),
286 	REF_CLK_FULL_DD(eip97),
287 	REF_CLK_PM_CPU(cpu),
288 	{ },
289 };
290 
291 /* SB periph clocks */
292 PERIPH_CLK_MUX_DD(gbe_50, 6, DIV_SEL2, DIV_SEL2, 6, 9);
293 PERIPH_CLK_MUX_DD(gbe_core, 8, DIV_SEL1, DIV_SEL1, 18, 21);
294 PERIPH_CLK_MUX_DD(gbe_125, 10, DIV_SEL1, DIV_SEL1, 6, 9);
295 static PERIPH_GATE(gbe1_50, 0);
296 static PERIPH_GATE(gbe0_50, 1);
297 static PERIPH_GATE(gbe1_125, 2);
298 static PERIPH_GATE(gbe0_125, 3);
299 PERIPH_CLK_GATE_DIV(gbe1_core, 4, DIV_SEL1, 13, clk_table1);
300 PERIPH_CLK_GATE_DIV(gbe0_core, 5, DIV_SEL1, 14, clk_table1);
301 PERIPH_CLK_GATE_DIV(gbe_bm, 12, DIV_SEL1, 0, clk_table1);
302 PERIPH_CLK_FULL_DD(sdio, 11, 14, DIV_SEL0, DIV_SEL0, 3, 6);
303 PERIPH_CLK_FULL_DD(usb32_usb2_sys, 16, 16, DIV_SEL0, DIV_SEL0, 9, 12);
304 PERIPH_CLK_FULL_DD(usb32_ss_sys, 17, 18, DIV_SEL0, DIV_SEL0, 15, 18);
305 
306 static struct clk_periph_data data_sb[] = {
307 	REF_CLK_MUX_DD(gbe_50),
308 	REF_CLK_MUX_DD(gbe_core),
309 	REF_CLK_MUX_DD(gbe_125),
310 	REF_CLK_GATE(gbe1_50, "gbe_50"),
311 	REF_CLK_GATE(gbe0_50, "gbe_50"),
312 	REF_CLK_GATE(gbe1_125, "gbe_125"),
313 	REF_CLK_GATE(gbe0_125, "gbe_125"),
314 	REF_CLK_GATE_DIV(gbe1_core, "gbe_core"),
315 	REF_CLK_GATE_DIV(gbe0_core, "gbe_core"),
316 	REF_CLK_GATE_DIV(gbe_bm, "gbe_core"),
317 	REF_CLK_FULL_DD(sdio),
318 	REF_CLK_FULL_DD(usb32_usb2_sys),
319 	REF_CLK_FULL_DD(usb32_ss_sys),
320 	{ },
321 };
322 
323 static unsigned int get_div(void __iomem *reg, int shift)
324 {
325 	u32 val;
326 
327 	val = (readl(reg) >> shift) & 0x7;
328 	if (val > 6)
329 		return 0;
330 	return val;
331 }
332 
333 static unsigned long clk_double_div_recalc_rate(struct clk_hw *hw,
334 						unsigned long parent_rate)
335 {
336 	struct clk_double_div *double_div = to_clk_double_div(hw);
337 	unsigned int div;
338 
339 	div = get_div(double_div->reg1, double_div->shift1);
340 	div *= get_div(double_div->reg2, double_div->shift2);
341 
342 	return DIV_ROUND_UP_ULL((u64)parent_rate, div);
343 }
344 
345 static const struct clk_ops clk_double_div_ops = {
346 	.recalc_rate = clk_double_div_recalc_rate,
347 };
348 
349 static void armada_3700_pm_dvfs_update_regs(unsigned int load_level,
350 					    unsigned int *reg,
351 					    unsigned int *offset)
352 {
353 	if (load_level <= ARMADA_37XX_DVFS_LOAD_1)
354 		*reg = ARMADA_37XX_NB_L0L1;
355 	else
356 		*reg = ARMADA_37XX_NB_L2L3;
357 
358 	if (load_level == ARMADA_37XX_DVFS_LOAD_0 ||
359 	    load_level ==  ARMADA_37XX_DVFS_LOAD_2)
360 		*offset += ARMADA_37XX_NB_CONFIG_SHIFT;
361 }
362 
363 static bool armada_3700_pm_dvfs_is_enabled(struct regmap *base)
364 {
365 	unsigned int val, reg = ARMADA_37XX_NB_DYN_MOD;
366 
367 	if (IS_ERR(base))
368 		return false;
369 
370 	regmap_read(base, reg, &val);
371 
372 	return !!(val & BIT(ARMADA_37XX_NB_DFS_EN));
373 }
374 
375 static unsigned int armada_3700_pm_dvfs_get_cpu_div(struct regmap *base)
376 {
377 	unsigned int reg = ARMADA_37XX_NB_CPU_LOAD;
378 	unsigned int offset = ARMADA_37XX_NB_TBG_DIV_OFF;
379 	unsigned int load_level, div;
380 
381 	/*
382 	 * This function is always called after the function
383 	 * armada_3700_pm_dvfs_is_enabled, so no need to check again
384 	 * if the base is valid.
385 	 */
386 	regmap_read(base, reg, &load_level);
387 
388 	/*
389 	 * The register and the offset inside this register accessed to
390 	 * read the current divider depend on the load level
391 	 */
392 	load_level &= ARMADA_37XX_NB_CPU_LOAD_MASK;
393 	armada_3700_pm_dvfs_update_regs(load_level, &reg, &offset);
394 
395 	regmap_read(base, reg, &div);
396 
397 	return (div >> offset) & ARMADA_37XX_NB_TBG_DIV_MASK;
398 }
399 
400 static unsigned int armada_3700_pm_dvfs_get_cpu_parent(struct regmap *base)
401 {
402 	unsigned int reg = ARMADA_37XX_NB_CPU_LOAD;
403 	unsigned int offset = ARMADA_37XX_NB_TBG_SEL_OFF;
404 	unsigned int load_level, sel;
405 
406 	/*
407 	 * This function is always called after the function
408 	 * armada_3700_pm_dvfs_is_enabled, so no need to check again
409 	 * if the base is valid
410 	 */
411 	regmap_read(base, reg, &load_level);
412 
413 	/*
414 	 * The register and the offset inside this register accessed to
415 	 * read the current divider depend on the load level
416 	 */
417 	load_level &= ARMADA_37XX_NB_CPU_LOAD_MASK;
418 	armada_3700_pm_dvfs_update_regs(load_level, &reg, &offset);
419 
420 	regmap_read(base, reg, &sel);
421 
422 	return (sel >> offset) & ARMADA_37XX_NB_TBG_SEL_MASK;
423 }
424 
425 static u8 clk_pm_cpu_get_parent(struct clk_hw *hw)
426 {
427 	struct clk_pm_cpu *pm_cpu = to_clk_pm_cpu(hw);
428 	u32 val;
429 
430 	if (armada_3700_pm_dvfs_is_enabled(pm_cpu->nb_pm_base)) {
431 		val = armada_3700_pm_dvfs_get_cpu_parent(pm_cpu->nb_pm_base);
432 	} else {
433 		val = readl(pm_cpu->reg_mux) >> pm_cpu->shift_mux;
434 		val &= pm_cpu->mask_mux;
435 	}
436 
437 	return val;
438 }
439 
440 static int clk_pm_cpu_set_parent(struct clk_hw *hw, u8 index)
441 {
442 	struct clk_pm_cpu *pm_cpu = to_clk_pm_cpu(hw);
443 	struct regmap *base = pm_cpu->nb_pm_base;
444 	int load_level;
445 
446 	/*
447 	 * We set the clock parent only if the DVFS is available but
448 	 * not enabled.
449 	 */
450 	if (IS_ERR(base) || armada_3700_pm_dvfs_is_enabled(base))
451 		return -EINVAL;
452 
453 	/* Set the parent clock for all the load level */
454 	for (load_level = 0; load_level < LOAD_LEVEL_NR; load_level++) {
455 		unsigned int reg, mask,  val,
456 			offset = ARMADA_37XX_NB_TBG_SEL_OFF;
457 
458 		armada_3700_pm_dvfs_update_regs(load_level, &reg, &offset);
459 
460 		val = index << offset;
461 		mask = ARMADA_37XX_NB_TBG_SEL_MASK << offset;
462 		regmap_update_bits(base, reg, mask, val);
463 	}
464 	return 0;
465 }
466 
467 static unsigned long clk_pm_cpu_recalc_rate(struct clk_hw *hw,
468 					    unsigned long parent_rate)
469 {
470 	struct clk_pm_cpu *pm_cpu = to_clk_pm_cpu(hw);
471 	unsigned int div;
472 
473 	if (armada_3700_pm_dvfs_is_enabled(pm_cpu->nb_pm_base))
474 		div = armada_3700_pm_dvfs_get_cpu_div(pm_cpu->nb_pm_base);
475 	else
476 		div = get_div(pm_cpu->reg_div, pm_cpu->shift_div);
477 	return DIV_ROUND_UP_ULL((u64)parent_rate, div);
478 }
479 
480 static long clk_pm_cpu_round_rate(struct clk_hw *hw, unsigned long rate,
481 				  unsigned long *parent_rate)
482 {
483 	struct clk_pm_cpu *pm_cpu = to_clk_pm_cpu(hw);
484 	struct regmap *base = pm_cpu->nb_pm_base;
485 	unsigned int div = *parent_rate / rate;
486 	unsigned int load_level;
487 	/* only available when DVFS is enabled */
488 	if (!armada_3700_pm_dvfs_is_enabled(base))
489 		return -EINVAL;
490 
491 	for (load_level = 0; load_level < LOAD_LEVEL_NR; load_level++) {
492 		unsigned int reg, val, offset = ARMADA_37XX_NB_TBG_DIV_OFF;
493 
494 		armada_3700_pm_dvfs_update_regs(load_level, &reg, &offset);
495 
496 		regmap_read(base, reg, &val);
497 
498 		val >>= offset;
499 		val &= ARMADA_37XX_NB_TBG_DIV_MASK;
500 		if (val == div)
501 			/*
502 			 * We found a load level matching the target
503 			 * divider, switch to this load level and
504 			 * return.
505 			 */
506 			return *parent_rate / div;
507 	}
508 
509 	/* We didn't find any valid divider */
510 	return -EINVAL;
511 }
512 
513 /*
514  * Switching the CPU from the L2 or L3 frequencies (300 and 200 Mhz
515  * respectively) to L0 frequency (1.2 Ghz) requires a significant
516  * amount of time to let VDD stabilize to the appropriate
517  * voltage. This amount of time is large enough that it cannot be
518  * covered by the hardware countdown register. Due to this, the CPU
519  * might start operating at L0 before the voltage is stabilized,
520  * leading to CPU stalls.
521  *
522  * To work around this problem, we prevent switching directly from the
523  * L2/L3 frequencies to the L0 frequency, and instead switch to the L1
524  * frequency in-between. The sequence therefore becomes:
525  * 1. First switch from L2/L3(200/300MHz) to L1(600MHZ)
526  * 2. Sleep 20ms for stabling VDD voltage
527  * 3. Then switch from L1(600MHZ) to L0(1200Mhz).
528  */
529 static void clk_pm_cpu_set_rate_wa(unsigned long rate, struct regmap *base)
530 {
531 	unsigned int cur_level;
532 
533 	if (rate != 1200 * 1000 * 1000)
534 		return;
535 
536 	regmap_read(base, ARMADA_37XX_NB_CPU_LOAD, &cur_level);
537 	cur_level &= ARMADA_37XX_NB_CPU_LOAD_MASK;
538 	if (cur_level <= ARMADA_37XX_DVFS_LOAD_1)
539 		return;
540 
541 	regmap_update_bits(base, ARMADA_37XX_NB_CPU_LOAD,
542 			   ARMADA_37XX_NB_CPU_LOAD_MASK,
543 			   ARMADA_37XX_DVFS_LOAD_1);
544 	msleep(20);
545 }
546 
547 static int clk_pm_cpu_set_rate(struct clk_hw *hw, unsigned long rate,
548 			       unsigned long parent_rate)
549 {
550 	struct clk_pm_cpu *pm_cpu = to_clk_pm_cpu(hw);
551 	struct regmap *base = pm_cpu->nb_pm_base;
552 	unsigned int div = parent_rate / rate;
553 	unsigned int load_level;
554 
555 	/* only available when DVFS is enabled */
556 	if (!armada_3700_pm_dvfs_is_enabled(base))
557 		return -EINVAL;
558 
559 	for (load_level = 0; load_level < LOAD_LEVEL_NR; load_level++) {
560 		unsigned int reg, mask, val,
561 			offset = ARMADA_37XX_NB_TBG_DIV_OFF;
562 
563 		armada_3700_pm_dvfs_update_regs(load_level, &reg, &offset);
564 
565 		regmap_read(base, reg, &val);
566 		val >>= offset;
567 		val &= ARMADA_37XX_NB_TBG_DIV_MASK;
568 
569 		if (val == div) {
570 			/*
571 			 * We found a load level matching the target
572 			 * divider, switch to this load level and
573 			 * return.
574 			 */
575 			reg = ARMADA_37XX_NB_CPU_LOAD;
576 			mask = ARMADA_37XX_NB_CPU_LOAD_MASK;
577 
578 			clk_pm_cpu_set_rate_wa(rate, base);
579 
580 			regmap_update_bits(base, reg, mask, load_level);
581 
582 			return rate;
583 		}
584 	}
585 
586 	/* We didn't find any valid divider */
587 	return -EINVAL;
588 }
589 
590 static const struct clk_ops clk_pm_cpu_ops = {
591 	.get_parent = clk_pm_cpu_get_parent,
592 	.set_parent = clk_pm_cpu_set_parent,
593 	.round_rate = clk_pm_cpu_round_rate,
594 	.set_rate = clk_pm_cpu_set_rate,
595 	.recalc_rate = clk_pm_cpu_recalc_rate,
596 };
597 
598 static const struct of_device_id armada_3700_periph_clock_of_match[] = {
599 	{ .compatible = "marvell,armada-3700-periph-clock-nb",
600 	  .data = data_nb, },
601 	{ .compatible = "marvell,armada-3700-periph-clock-sb",
602 	.data = data_sb, },
603 	{ }
604 };
605 
606 static int armada_3700_add_composite_clk(const struct clk_periph_data *data,
607 					 void __iomem *reg, spinlock_t *lock,
608 					 struct device *dev, struct clk_hw **hw)
609 {
610 	const struct clk_ops *mux_ops = NULL, *gate_ops = NULL,
611 		*rate_ops = NULL;
612 	struct clk_hw *mux_hw = NULL, *gate_hw = NULL, *rate_hw = NULL;
613 
614 	if (data->mux_hw) {
615 		struct clk_mux *mux;
616 
617 		mux_hw = data->mux_hw;
618 		mux = to_clk_mux(mux_hw);
619 		mux->lock = lock;
620 		mux_ops = mux_hw->init->ops;
621 		mux->reg = reg + (u64)mux->reg;
622 	}
623 
624 	if (data->gate_hw) {
625 		struct clk_gate *gate;
626 
627 		gate_hw = data->gate_hw;
628 		gate = to_clk_gate(gate_hw);
629 		gate->lock = lock;
630 		gate_ops = gate_hw->init->ops;
631 		gate->reg = reg + (u64)gate->reg;
632 		gate->flags = CLK_GATE_SET_TO_DISABLE;
633 	}
634 
635 	if (data->rate_hw) {
636 		rate_hw = data->rate_hw;
637 		rate_ops = rate_hw->init->ops;
638 		if (data->is_double_div) {
639 			struct clk_double_div *rate;
640 
641 			rate =  to_clk_double_div(rate_hw);
642 			rate->reg1 = reg + (u64)rate->reg1;
643 			rate->reg2 = reg + (u64)rate->reg2;
644 		} else {
645 			struct clk_divider *rate = to_clk_divider(rate_hw);
646 			const struct clk_div_table *clkt;
647 			int table_size = 0;
648 
649 			rate->reg = reg + (u64)rate->reg;
650 			for (clkt = rate->table; clkt->div; clkt++)
651 				table_size++;
652 			rate->width = order_base_2(table_size);
653 			rate->lock = lock;
654 		}
655 	}
656 
657 	if (data->muxrate_hw) {
658 		struct clk_pm_cpu *pmcpu_clk;
659 		struct clk_hw *muxrate_hw = data->muxrate_hw;
660 		struct regmap *map;
661 
662 		pmcpu_clk =  to_clk_pm_cpu(muxrate_hw);
663 		pmcpu_clk->reg_mux = reg + (u64)pmcpu_clk->reg_mux;
664 		pmcpu_clk->reg_div = reg + (u64)pmcpu_clk->reg_div;
665 
666 		mux_hw = muxrate_hw;
667 		rate_hw = muxrate_hw;
668 		mux_ops = muxrate_hw->init->ops;
669 		rate_ops = muxrate_hw->init->ops;
670 
671 		map = syscon_regmap_lookup_by_compatible(
672 				"marvell,armada-3700-nb-pm");
673 		pmcpu_clk->nb_pm_base = map;
674 	}
675 
676 	*hw = clk_hw_register_composite(dev, data->name, data->parent_names,
677 					data->num_parents, mux_hw,
678 					mux_ops, rate_hw, rate_ops,
679 					gate_hw, gate_ops, CLK_IGNORE_UNUSED);
680 
681 	return PTR_ERR_OR_ZERO(*hw);
682 }
683 
684 static int __maybe_unused armada_3700_periph_clock_suspend(struct device *dev)
685 {
686 	struct clk_periph_driver_data *data = dev_get_drvdata(dev);
687 
688 	data->tbg_sel = readl(data->reg + TBG_SEL);
689 	data->div_sel0 = readl(data->reg + DIV_SEL0);
690 	data->div_sel1 = readl(data->reg + DIV_SEL1);
691 	data->div_sel2 = readl(data->reg + DIV_SEL2);
692 	data->clk_sel = readl(data->reg + CLK_SEL);
693 	data->clk_dis = readl(data->reg + CLK_DIS);
694 
695 	return 0;
696 }
697 
698 static int __maybe_unused armada_3700_periph_clock_resume(struct device *dev)
699 {
700 	struct clk_periph_driver_data *data = dev_get_drvdata(dev);
701 
702 	/* Follow the same order than what the Cortex-M3 does (ATF code) */
703 	writel(data->clk_dis, data->reg + CLK_DIS);
704 	writel(data->div_sel0, data->reg + DIV_SEL0);
705 	writel(data->div_sel1, data->reg + DIV_SEL1);
706 	writel(data->div_sel2, data->reg + DIV_SEL2);
707 	writel(data->tbg_sel, data->reg + TBG_SEL);
708 	writel(data->clk_sel, data->reg + CLK_SEL);
709 
710 	return 0;
711 }
712 
713 static const struct dev_pm_ops armada_3700_periph_clock_pm_ops = {
714 	SET_SYSTEM_SLEEP_PM_OPS(armada_3700_periph_clock_suspend,
715 				armada_3700_periph_clock_resume)
716 };
717 
718 static int armada_3700_periph_clock_probe(struct platform_device *pdev)
719 {
720 	struct clk_periph_driver_data *driver_data;
721 	struct device_node *np = pdev->dev.of_node;
722 	const struct clk_periph_data *data;
723 	struct device *dev = &pdev->dev;
724 	int num_periph = 0, i, ret;
725 	struct resource *res;
726 
727 	data = of_device_get_match_data(dev);
728 	if (!data)
729 		return -ENODEV;
730 
731 	while (data[num_periph].name)
732 		num_periph++;
733 
734 	driver_data = devm_kzalloc(dev, sizeof(*driver_data), GFP_KERNEL);
735 	if (!driver_data)
736 		return -ENOMEM;
737 
738 	driver_data->hw_data = devm_kzalloc(dev,
739 					    struct_size(driver_data->hw_data,
740 							hws, num_periph),
741 					    GFP_KERNEL);
742 	if (!driver_data->hw_data)
743 		return -ENOMEM;
744 	driver_data->hw_data->num = num_periph;
745 
746 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
747 	driver_data->reg = devm_ioremap_resource(dev, res);
748 	if (IS_ERR(driver_data->reg))
749 		return PTR_ERR(driver_data->reg);
750 
751 	spin_lock_init(&driver_data->lock);
752 
753 	for (i = 0; i < num_periph; i++) {
754 		struct clk_hw **hw = &driver_data->hw_data->hws[i];
755 		if (armada_3700_add_composite_clk(&data[i], driver_data->reg,
756 						  &driver_data->lock, dev, hw))
757 			dev_err(dev, "Can't register periph clock %s\n",
758 				data[i].name);
759 	}
760 
761 	ret = of_clk_add_hw_provider(np, of_clk_hw_onecell_get,
762 				     driver_data->hw_data);
763 	if (ret) {
764 		for (i = 0; i < num_periph; i++)
765 			clk_hw_unregister(driver_data->hw_data->hws[i]);
766 		return ret;
767 	}
768 
769 	platform_set_drvdata(pdev, driver_data);
770 	return 0;
771 }
772 
773 static int armada_3700_periph_clock_remove(struct platform_device *pdev)
774 {
775 	struct clk_periph_driver_data *data = platform_get_drvdata(pdev);
776 	struct clk_hw_onecell_data *hw_data = data->hw_data;
777 	int i;
778 
779 	of_clk_del_provider(pdev->dev.of_node);
780 
781 	for (i = 0; i < hw_data->num; i++)
782 		clk_hw_unregister(hw_data->hws[i]);
783 
784 	return 0;
785 }
786 
787 static struct platform_driver armada_3700_periph_clock_driver = {
788 	.probe = armada_3700_periph_clock_probe,
789 	.remove = armada_3700_periph_clock_remove,
790 	.driver		= {
791 		.name	= "marvell-armada-3700-periph-clock",
792 		.of_match_table = armada_3700_periph_clock_of_match,
793 		.pm	= &armada_3700_periph_clock_pm_ops,
794 	},
795 };
796 
797 builtin_platform_driver(armada_3700_periph_clock_driver);
798