1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2019-20 Sean Anderson <seanga2@gmail.com>
5 #define LOG_CATEGORY UCLASS_CLK
9 #include <clk-uclass.h>
15 #include <dt-bindings/clock/k210-sysctl.h>
16 #include <dt-bindings/mfd/k210-sysctl.h>
18 #include <linux/bitfield.h>
20 DECLARE_GLOBAL_DATA_PTR;
23 * struct k210_clk_priv - K210 clock driver private data
24 * @base: The base address of the sysctl device
25 * @in0: The "in0" external oscillator
27 struct k210_clk_priv {
33 * All parameters for different sub-clocks are collected into parameter arrays.
34 * These parameters are then initialized by the clock which uses them during
35 * probe. To save space, ids are automatically generated for each sub-clock by
36 * using an enum. Instead of storing a parameter struct for each clock, even for
37 * those clocks which don't use a particular type of sub-clock, we can just
38 * store the parameters for the clocks which need them.
40 * So why do it like this? Arranging all the sub-clocks together makes it very
41 * easy to find bugs in the code.
45 * enum k210_clk_div_type - The type of divider
46 * @K210_DIV_ONE: freq = parent / (reg + 1)
47 * @K210_DIV_EVEN: freq = parent / 2 / (reg + 1)
48 * @K210_DIV_POWER: freq = parent / (2 << reg)
49 * @K210_DIV_FIXED: freq = parent / factor
51 enum k210_clk_div_type {
59 * struct k210_div_params - Parameters for dividing clocks
60 * @type: An &enum k210_clk_div_type specifying the dividing formula
61 * @off: The offset of the divider from the sysctl base address
62 * @shift: The offset of the LSB of the divider
63 * @width: The number of bits in the divider
64 * @div: The fixed divisor for this divider
66 struct k210_div_params {
79 DIV(K210_CLK_ACLK, K210_SYSCTL_SEL0, 1, 2, K210_DIV_POWER) \
80 DIV(K210_CLK_APB0, K210_SYSCTL_SEL0, 3, 3, K210_DIV_ONE) \
81 DIV(K210_CLK_APB1, K210_SYSCTL_SEL0, 6, 3, K210_DIV_ONE) \
82 DIV(K210_CLK_APB2, K210_SYSCTL_SEL0, 9, 3, K210_DIV_ONE) \
83 DIV(K210_CLK_SRAM0, K210_SYSCTL_THR0, 0, 4, K210_DIV_ONE) \
84 DIV(K210_CLK_SRAM1, K210_SYSCTL_THR0, 4, 4, K210_DIV_ONE) \
85 DIV(K210_CLK_AI, K210_SYSCTL_THR0, 8, 4, K210_DIV_ONE) \
86 DIV(K210_CLK_DVP, K210_SYSCTL_THR0, 12, 4, K210_DIV_ONE) \
87 DIV(K210_CLK_ROM, K210_SYSCTL_THR0, 16, 4, K210_DIV_ONE) \
88 DIV(K210_CLK_SPI0, K210_SYSCTL_THR1, 0, 8, K210_DIV_EVEN) \
89 DIV(K210_CLK_SPI1, K210_SYSCTL_THR1, 8, 8, K210_DIV_EVEN) \
90 DIV(K210_CLK_SPI2, K210_SYSCTL_THR1, 16, 8, K210_DIV_EVEN) \
91 DIV(K210_CLK_SPI3, K210_SYSCTL_THR1, 24, 8, K210_DIV_EVEN) \
92 DIV(K210_CLK_TIMER0, K210_SYSCTL_THR2, 0, 8, K210_DIV_EVEN) \
93 DIV(K210_CLK_TIMER1, K210_SYSCTL_THR2, 8, 8, K210_DIV_EVEN) \
94 DIV(K210_CLK_TIMER2, K210_SYSCTL_THR2, 16, 8, K210_DIV_EVEN) \
95 DIV(K210_CLK_I2S0, K210_SYSCTL_THR3, 0, 16, K210_DIV_EVEN) \
96 DIV(K210_CLK_I2S1, K210_SYSCTL_THR3, 16, 16, K210_DIV_EVEN) \
97 DIV(K210_CLK_I2S2, K210_SYSCTL_THR4, 0, 16, K210_DIV_EVEN) \
98 DIV(K210_CLK_I2S0_M, K210_SYSCTL_THR4, 16, 8, K210_DIV_EVEN) \
99 DIV(K210_CLK_I2S1_M, K210_SYSCTL_THR4, 24, 8, K210_DIV_EVEN) \
100 DIV(K210_CLK_I2S2_M, K210_SYSCTL_THR4, 0, 8, K210_DIV_EVEN) \
101 DIV(K210_CLK_I2C0, K210_SYSCTL_THR5, 8, 8, K210_DIV_EVEN) \
102 DIV(K210_CLK_I2C1, K210_SYSCTL_THR5, 16, 8, K210_DIV_EVEN) \
103 DIV(K210_CLK_I2C2, K210_SYSCTL_THR5, 24, 8, K210_DIV_EVEN) \
104 DIV(K210_CLK_WDT0, K210_SYSCTL_THR6, 0, 8, K210_DIV_EVEN) \
105 DIV(K210_CLK_WDT1, K210_SYSCTL_THR6, 8, 8, K210_DIV_EVEN) \
106 DIV_FIXED(K210_CLK_CLINT, 50) \
108 #define _DIVIFY(id) K210_CLK_DIV_##id
109 #define DIVIFY(id) _DIVIFY(id)
112 #define DIV(id, ...) DIVIFY(id),
113 #define DIV_FIXED DIV
120 static const struct k210_div_params k210_divs[] = {
121 #define DIV(id, _off, _shift, _width, _type) \
128 #define DIV_FIXED(id, _div) \
130 .type = K210_DIV_FIXED, \
142 * struct k210_gate_params - Parameters for gated clocks
143 * @off: The offset of the gate from the sysctl base address
144 * @bit_idx: The index of the bit within the register
146 struct k210_gate_params {
152 GATE(K210_CLK_CPU, K210_SYSCTL_EN_CENT, 0) \
153 GATE(K210_CLK_SRAM0, K210_SYSCTL_EN_CENT, 1) \
154 GATE(K210_CLK_SRAM1, K210_SYSCTL_EN_CENT, 2) \
155 GATE(K210_CLK_APB0, K210_SYSCTL_EN_CENT, 3) \
156 GATE(K210_CLK_APB1, K210_SYSCTL_EN_CENT, 4) \
157 GATE(K210_CLK_APB2, K210_SYSCTL_EN_CENT, 5) \
158 GATE(K210_CLK_ROM, K210_SYSCTL_EN_PERI, 0) \
159 GATE(K210_CLK_DMA, K210_SYSCTL_EN_PERI, 1) \
160 GATE(K210_CLK_AI, K210_SYSCTL_EN_PERI, 2) \
161 GATE(K210_CLK_DVP, K210_SYSCTL_EN_PERI, 3) \
162 GATE(K210_CLK_FFT, K210_SYSCTL_EN_PERI, 4) \
163 GATE(K210_CLK_GPIO, K210_SYSCTL_EN_PERI, 5) \
164 GATE(K210_CLK_SPI0, K210_SYSCTL_EN_PERI, 6) \
165 GATE(K210_CLK_SPI1, K210_SYSCTL_EN_PERI, 7) \
166 GATE(K210_CLK_SPI2, K210_SYSCTL_EN_PERI, 8) \
167 GATE(K210_CLK_SPI3, K210_SYSCTL_EN_PERI, 9) \
168 GATE(K210_CLK_I2S0, K210_SYSCTL_EN_PERI, 10) \
169 GATE(K210_CLK_I2S1, K210_SYSCTL_EN_PERI, 11) \
170 GATE(K210_CLK_I2S2, K210_SYSCTL_EN_PERI, 12) \
171 GATE(K210_CLK_I2C0, K210_SYSCTL_EN_PERI, 13) \
172 GATE(K210_CLK_I2C1, K210_SYSCTL_EN_PERI, 14) \
173 GATE(K210_CLK_I2C2, K210_SYSCTL_EN_PERI, 15) \
174 GATE(K210_CLK_UART1, K210_SYSCTL_EN_PERI, 16) \
175 GATE(K210_CLK_UART2, K210_SYSCTL_EN_PERI, 17) \
176 GATE(K210_CLK_UART3, K210_SYSCTL_EN_PERI, 18) \
177 GATE(K210_CLK_AES, K210_SYSCTL_EN_PERI, 19) \
178 GATE(K210_CLK_FPIOA, K210_SYSCTL_EN_PERI, 20) \
179 GATE(K210_CLK_TIMER0, K210_SYSCTL_EN_PERI, 21) \
180 GATE(K210_CLK_TIMER1, K210_SYSCTL_EN_PERI, 22) \
181 GATE(K210_CLK_TIMER2, K210_SYSCTL_EN_PERI, 23) \
182 GATE(K210_CLK_WDT0, K210_SYSCTL_EN_PERI, 24) \
183 GATE(K210_CLK_WDT1, K210_SYSCTL_EN_PERI, 25) \
184 GATE(K210_CLK_SHA, K210_SYSCTL_EN_PERI, 26) \
185 GATE(K210_CLK_OTP, K210_SYSCTL_EN_PERI, 27) \
186 GATE(K210_CLK_RTC, K210_SYSCTL_EN_PERI, 29)
188 #define _GATEIFY(id) K210_CLK_GATE_##id
189 #define GATEIFY(id) _GATEIFY(id)
192 #define GATE(id, ...) GATEIFY(id),
198 static const struct k210_gate_params k210_gates[] = {
199 #define GATE(id, _off, _idx) \
210 /* The most parents is PLL2 */
211 #define K210_CLK_MAX_PARENTS 3
214 * struct k210_mux_params - Parameters for muxed clocks
215 * @parents: A list of parent clock ids
216 * @num_parents: The number of parent clocks
217 * @off: The offset of the mux from the base sysctl address
218 * @shift: The offset of the LSB of the mux selector
219 * @width: The number of bits in the mux selector
221 struct k210_mux_params {
222 u8 parents[K210_CLK_MAX_PARENTS];
229 #define MUX(id, reg, shift, width) \
230 MUX_PARENTS(id, reg, shift, width, K210_CLK_IN0, K210_CLK_PLL0)
232 MUX_PARENTS(K210_CLK_PLL2, K210_SYSCTL_PLL2, 26, 2, \
233 K210_CLK_IN0, K210_CLK_PLL0, K210_CLK_PLL1) \
234 MUX(K210_CLK_ACLK, K210_SYSCTL_SEL0, 0, 1) \
235 MUX(K210_CLK_SPI3, K210_SYSCTL_SEL0, 12, 1) \
236 MUX(K210_CLK_TIMER0, K210_SYSCTL_SEL0, 13, 1) \
237 MUX(K210_CLK_TIMER1, K210_SYSCTL_SEL0, 14, 1) \
238 MUX(K210_CLK_TIMER2, K210_SYSCTL_SEL0, 15, 1)
240 #define _MUXIFY(id) K210_CLK_MUX_##id
241 #define MUXIFY(id) _MUXIFY(id)
244 #define MUX_PARENTS(id, ...) MUXIFY(id),
250 static const struct k210_mux_params k210_muxes[] = {
251 #define MUX_PARENTS(id, _off, _shift, _width, ...) \
253 .parents = { __VA_ARGS__ }, \
254 .num_parents = __count_args(__VA_ARGS__), \
267 * struct k210_pll_params - K210 PLL parameters
268 * @off: The offset of the PLL from the base sysctl address
269 * @shift: The offset of the LSB of the lock status
270 * @width: The number of bits in the lock status
272 struct k210_pll_params {
278 static const struct k210_pll_params k210_plls[] = {
279 #define PLL(_off, _shift, _width) { \
284 [0] = PLL(K210_SYSCTL_PLL0, 0, 2),
285 [1] = PLL(K210_SYSCTL_PLL1, 8, 1),
286 [2] = PLL(K210_SYSCTL_PLL2, 16, 1),
291 * enum k210_clk_flags - The type of a K210 clock
292 * @K210_CLKF_MUX: This clock has a mux and not a static parent
293 * @K210_CLKF_PLL: This clock is a PLL
295 enum k210_clk_flags {
296 K210_CLKF_MUX = BIT(0),
297 K210_CLKF_PLL = BIT(1),
301 * struct k210_clk_params - The parameters defining a K210 clock
302 * @name: The name of the clock
303 * @flags: A set of &enum k210_clk_flags defining which fields are valid
304 * @mux: An &enum k210_mux_id of this clock's mux
305 * @parent: The clock id of this clock's parent
306 * @pll: The id of the PLL (if this clock is a PLL)
307 * @div: An &enum k210_div_id of this clock's divider
308 * @gate: An &enum k210_gate_id of this clock's gate
310 struct k210_clk_params {
311 #if IS_ENABLED(CONFIG_CMD_CLK)
328 static const struct k210_clk_params k210_clks[] = {
329 #if IS_ENABLED(CONFIG_CMD_CLK)
330 #define NAME(_name) .name = (_name),
334 #define CLK(id, _name, _parent, _div, _gate) \
337 .parent = (_parent), \
341 #define CLK_MUX(id, _name, _mux, _div, _gate) \
344 .flags = K210_CLKF_MUX, \
349 #define CLK_PLL(id, _pll, _parent) \
352 .flags = K210_CLKF_PLL, \
353 .parent = (_parent), \
356 #define CLK_FULL(id, name) \
357 CLK_MUX(id, name, MUXIFY(id), DIVIFY(id), GATEIFY(id))
358 #define CLK_NOMUX(id, name, parent) \
359 CLK(id, name, parent, DIVIFY(id), GATEIFY(id))
360 #define CLK_DIV(id, name, parent) \
361 CLK(id, name, parent, DIVIFY(id), K210_CLK_GATE_NONE)
362 #define CLK_GATE(id, name, parent) \
363 CLK(id, name, parent, K210_CLK_DIV_NONE, GATEIFY(id))
364 CLK_PLL(K210_CLK_PLL0, 0, K210_CLK_IN0),
365 CLK_PLL(K210_CLK_PLL1, 1, K210_CLK_IN0),
368 .flags = K210_CLKF_MUX | K210_CLKF_PLL,
369 .mux = MUXIFY(K210_CLK_PLL2),
372 CLK_MUX(K210_CLK_ACLK, "aclk", MUXIFY(K210_CLK_ACLK),
373 DIVIFY(K210_CLK_ACLK), K210_CLK_GATE_NONE),
374 CLK_FULL(K210_CLK_SPI3, "spi3"),
375 CLK_FULL(K210_CLK_TIMER0, "timer0"),
376 CLK_FULL(K210_CLK_TIMER1, "timer1"),
377 CLK_FULL(K210_CLK_TIMER2, "timer2"),
378 CLK_NOMUX(K210_CLK_SRAM0, "sram0", K210_CLK_ACLK),
379 CLK_NOMUX(K210_CLK_SRAM1, "sram1", K210_CLK_ACLK),
380 CLK_NOMUX(K210_CLK_ROM, "rom", K210_CLK_ACLK),
381 CLK_NOMUX(K210_CLK_DVP, "dvp", K210_CLK_ACLK),
382 CLK_NOMUX(K210_CLK_APB0, "apb0", K210_CLK_ACLK),
383 CLK_NOMUX(K210_CLK_APB1, "apb1", K210_CLK_ACLK),
384 CLK_NOMUX(K210_CLK_APB2, "apb2", K210_CLK_ACLK),
385 CLK_NOMUX(K210_CLK_AI, "ai", K210_CLK_PLL1),
386 CLK_NOMUX(K210_CLK_I2S0, "i2s0", K210_CLK_PLL2),
387 CLK_NOMUX(K210_CLK_I2S1, "i2s1", K210_CLK_PLL2),
388 CLK_NOMUX(K210_CLK_I2S2, "i2s2", K210_CLK_PLL2),
389 CLK_NOMUX(K210_CLK_WDT0, "wdt0", K210_CLK_IN0),
390 CLK_NOMUX(K210_CLK_WDT1, "wdt1", K210_CLK_IN0),
391 CLK_NOMUX(K210_CLK_SPI0, "spi0", K210_CLK_PLL0),
392 CLK_NOMUX(K210_CLK_SPI1, "spi1", K210_CLK_PLL0),
393 CLK_NOMUX(K210_CLK_SPI2, "spi2", K210_CLK_PLL0),
394 CLK_NOMUX(K210_CLK_I2C0, "i2c0", K210_CLK_PLL0),
395 CLK_NOMUX(K210_CLK_I2C1, "i2c1", K210_CLK_PLL0),
396 CLK_NOMUX(K210_CLK_I2C2, "i2c2", K210_CLK_PLL0),
397 CLK_DIV(K210_CLK_I2S0_M, "i2s0_m", K210_CLK_PLL2),
398 CLK_DIV(K210_CLK_I2S1_M, "i2s1_m", K210_CLK_PLL2),
399 CLK_DIV(K210_CLK_I2S2_M, "i2s2_m", K210_CLK_PLL2),
400 CLK_DIV(K210_CLK_CLINT, "clint", K210_CLK_ACLK),
401 CLK_GATE(K210_CLK_CPU, "cpu", K210_CLK_ACLK),
402 CLK_GATE(K210_CLK_DMA, "dma", K210_CLK_ACLK),
403 CLK_GATE(K210_CLK_FFT, "fft", K210_CLK_ACLK),
404 CLK_GATE(K210_CLK_GPIO, "gpio", K210_CLK_APB0),
405 CLK_GATE(K210_CLK_UART1, "uart1", K210_CLK_APB0),
406 CLK_GATE(K210_CLK_UART2, "uart2", K210_CLK_APB0),
407 CLK_GATE(K210_CLK_UART3, "uart3", K210_CLK_APB0),
408 CLK_GATE(K210_CLK_FPIOA, "fpioa", K210_CLK_APB0),
409 CLK_GATE(K210_CLK_SHA, "sha", K210_CLK_APB0),
410 CLK_GATE(K210_CLK_AES, "aes", K210_CLK_APB1),
411 CLK_GATE(K210_CLK_OTP, "otp", K210_CLK_APB1),
412 CLK_GATE(K210_CLK_RTC, "rtc", K210_CLK_IN0),
423 #define K210_PLL_CLKR GENMASK(3, 0)
424 #define K210_PLL_CLKF GENMASK(9, 4)
425 #define K210_PLL_CLKOD GENMASK(13, 10) /* Output Divider */
426 #define K210_PLL_BWADJ GENMASK(19, 14) /* BandWidth Adjust */
427 #define K210_PLL_RESET BIT(20)
428 #define K210_PLL_PWRD BIT(21) /* PoWeReD */
429 #define K210_PLL_INTFB BIT(22) /* Internal FeedBack */
430 #define K210_PLL_BYPASS BIT(23)
431 #define K210_PLL_TEST BIT(24)
432 #define K210_PLL_EN BIT(25)
433 #define K210_PLL_TEST_EN BIT(26)
435 #define K210_PLL_LOCK 0
436 #define K210_PLL_CLEAR_SLIP 2
437 #define K210_PLL_TEST_OUT 3
439 #ifdef CONFIG_CLK_K210_SET_RATE
440 static int k210_pll_enable(struct k210_clk_priv *priv, int id);
441 static int k210_pll_disable(struct k210_clk_priv *priv, int id);
442 static ulong k210_pll_get_rate(struct k210_clk_priv *priv, int id, ulong rate_in);
445 * The PLL included with the Kendryte K210 appears to be a True Circuits, Inc.
446 * General-Purpose PLL. The logical layout of the PLL with internal feedback is
447 * approximately the following:
465 * |phase detector|<---+
469 * +---+ |feedback clock|
470 * |VCO| +--------------+
485 * The k210 PLLs have three factors: r, f, and od. Because of the feedback mode,
486 * the effect of the division by f is to multiply the input frequency. The
487 * equation for the output rate is
488 * rate = (rate_in * f) / (r * od).
489 * Moving knowns to one side of the equation, we get
490 * rate / rate_in = f / (r * od)
491 * Rearranging slightly,
492 * abs_error = abs((rate / rate_in) - (f / (r * od))).
493 * To get relative, error, we divide by the expected ratio
494 * error = abs((rate / rate_in) - (f / (r * od))) / (rate / rate_in).
496 * error = abs(1 - f / (r * od)) / (rate / rate_in)
497 * error = abs(1 - (f * rate_in) / (r * od * rate))
498 * Using the constants ratio = rate / rate_in and inv_ratio = rate_in / rate,
499 * error = abs((f * inv_ratio) / (r * od) - 1)
500 * This is the error used in evaluating parameters.
502 * r and od are four bits each, while f is six bits. Because r and od are
503 * multiplied together, instead of the full 256 values possible if both bits
504 * were used fully, there are only 97 distinct products. Combined with f, there
505 * are 6208 theoretical settings for the PLL. However, most of these settings
506 * can be ruled out immediately because they do not have the correct ratio.
508 * In addition to the constraint of approximating the desired ratio, parameters
509 * must also keep internal pll frequencies within acceptable ranges. The divided
510 * clock's minimum and maximum frequencies have a ratio of around 128. This
511 * leaves fairly substantial room to work with, especially since the only
512 * affected parameter is r. The VCO's minimum and maximum frequency have a ratio
513 * of 5, which is considerably more restrictive.
515 * The r and od factors are stored in a table. This is to make it easy to find
516 * the next-largest product. Some products have multiple factorizations, but
517 * only when one factor has at least a 2.5x ratio to the factors of the other
518 * factorization. This is because any smaller ratio would not make a difference
519 * when ensuring the VCO's frequency is within spec.
521 * Throughout the calculation function, fixed point arithmetic is used. Because
522 * the range of rate and rate_in may be up to 1.75 GHz, or around 2^30, 64-bit
523 * 32.32 fixed-point numbers are used to represent ratios. In general, to
524 * implement division, the numerator is first multiplied by 2^32. This gives a
525 * result where the whole number part is in the upper 32 bits, and the fraction
526 * is in the lower 32 bits.
528 * In general, rounding is done to the closest integer. This helps find the best
529 * approximation for the ratio. Rounding in one direction (e.g down) could cause
530 * the function to miss a better ratio with one of the parameters increased by
535 * The factors table was generated with the following python code:
538 * return (1.0*x/y > 2.5) or (1.0*y/x > 2.5)
541 * for i in range(1, 17):
542 * for j in range(1, 17):
543 * fs = factors.get(i*j) or []
544 * if fs == [] or all([
545 * (p(i, x) and p(i, y)) or (p(j, x) and p(j, y))
546 * for (x, y) in fs]):
550 * for k, l in sorted(factors.items()):
552 * print("PACK(%s, %s)," % v)
554 #define PACK(r, od) (((((r) - 1) & 0xF) << 4) | (((od) - 1) & 0xF))
555 #define UNPACK_R(val) ((((val) >> 4) & 0xF) + 1)
556 #define UNPACK_OD(val) (((val) & 0xF) + 1)
557 static const u8 factors[] = {
661 TEST_STATIC int k210_pll_calc_config(u32 rate, u32 rate_in,
662 struct k210_pll_config *best)
665 s64 error, best_error;
666 u64 ratio, inv_ratio; /* fixed point 32.32 ratio of the rates */
671 * Can't go over 1.75 GHz or under 21.25 MHz due to limitations on the
672 * VCO frequency. These are not the same limits as below because od can
673 * reduce the output frequency by 16.
675 if (rate > 1750000000 || rate < 21250000)
678 /* Similar restrictions on the input rate */
679 if (rate_in > 1750000000 || rate_in < 13300000)
682 ratio = DIV_ROUND_CLOSEST_ULL((u64)rate << 32, rate_in);
683 inv_ratio = DIV_ROUND_CLOSEST_ULL((u64)rate_in << 32, rate);
684 /* Can't increase by more than 64 or reduce by more than 256 */
685 if (rate > rate_in && ratio > (64ULL << 32))
687 else if (rate <= rate_in && inv_ratio > (256ULL << 32))
691 * The divided clock (rate_in / r) must stay between 1.75 GHz and 13.3
692 * MHz. There is no minimum, since the only way to get a higher input
693 * clock than 26 MHz is to use a clock generated by a PLL. Because PLLs
694 * cannot output frequencies greater than 1.75 GHz, the minimum would
695 * never be greater than one.
697 max_r = DIV_ROUND_DOWN_ULL(rate_in, 13300000);
699 /* Variables get immediately incremented, so start at -1th iteration */
704 best_error = S64_MAX;
706 /* do-while here so we always try at least one ratio */
709 * Whether we swapped r and od while enforcing frequency limits
711 bool swapped = false;
713 * Whether the intermediate frequencies are out-of-spec
720 * Try the next largest value for f (or r and od) and
721 * recalculate the other parameters based on that
723 if (rate > rate_in) {
725 * Skip factors of the same product if we already tried
730 r = UNPACK_R(factors[i]);
731 od = UNPACK_OD(factors[i]);
732 } while (i + 1 < ARRAY_SIZE(factors) &&
733 r * od == last_r * last_od);
736 f = (r * od * ratio + BIT(31)) >> 32;
740 u64 tmp = ++f * inv_ratio;
741 bool round_up = !!(tmp & BIT(31));
742 u32 goal = (tmp >> 32) + round_up;
745 /* Get the next r/od pair in factors */
746 while (r * od < goal && i + 1 < ARRAY_SIZE(factors)) {
748 r = UNPACK_R(factors[i]);
749 od = UNPACK_OD(factors[i]);
753 * This is a case of double rounding. If we rounded up
754 * above, we need to round down (in cases of ties) here.
755 * This prevents off-by-one errors resulting from
756 * choosing X+2 over X when X.Y rounds up to X+1 and
757 * there is no r * od = X+1. For the converse, when X.Y
758 * is rounded down to X, we should choose X+1 over X-1.
760 err = abs(r * od - goal);
761 last_err = abs(last_r * last_od - goal);
762 if (last_err < err || (round_up && last_err == err)) {
770 * Enforce limits on internal clock frequencies. If we
771 * aren't in spec, try swapping r and od. If everything is
772 * in-spec, calculate the relative error.
780 * There is no way to only divide once; we need
781 * to examine the frequency with and without the
784 u64 vco = DIV_ROUND_CLOSEST_ULL(rate_in * f, r);
786 if (vco > 1750000000 || vco < 340000000)
803 * Try looking ahead to see if there are additional
804 * factors for the same product.
806 if (i + 1 < ARRAY_SIZE(factors)) {
808 new_r = UNPACK_R(factors[i]);
809 new_od = UNPACK_OD(factors[i]);
810 if (r * od == new_r * new_od) {
820 * Try looking back to see if there is a worse ratio
821 * that we could try anyway
825 new_r = UNPACK_R(factors[i]);
826 new_od = UNPACK_OD(factors[i]);
828 * Don't loop over factors for the same product
829 * to avoid getting stuck because of the above
832 if (r * od != new_r * new_od) {
833 if (new_r * new_od > last_r * last_od) {
843 /* We ran out of things to try */
847 error = DIV_ROUND_CLOSEST_ULL(f * inv_ratio, r * od);
848 /* The lower 16 bits are spurious */
849 error = abs64((error - BIT_ULL(32))) >> 16;
851 if (error < best_error) {
857 } while (f < 64 && i + 1 < ARRAY_SIZE(factors) && error != 0);
859 log_debug("best error %lld\n", best_error);
860 if (best_error == S64_MAX)
866 static ulong k210_pll_set_rate(struct k210_clk_priv *priv, int id, ulong rate,
870 const struct k210_pll_params *pll = &k210_plls[id];
871 struct k210_pll_config config = {};
875 err = k210_pll_calc_config(rate, rate_in, &config);
878 log_debug("Got r=%u f=%u od=%u\n", config.r, config.f, config.od);
880 /* Don't bother setting the rate if we're already at that rate */
881 calc_rate = DIV_ROUND_DOWN_ULL(((u64)rate_in) * config.f,
882 config.r * config.od);
883 if (calc_rate == k210_pll_get_rate(priv, id, rate))
886 k210_pll_disable(priv, id);
888 reg = readl(priv->base + pll->off);
889 reg &= ~K210_PLL_CLKR
893 reg |= FIELD_PREP(K210_PLL_CLKR, config.r - 1)
894 | FIELD_PREP(K210_PLL_CLKF, config.f - 1)
895 | FIELD_PREP(K210_PLL_CLKOD, config.od - 1)
896 | FIELD_PREP(K210_PLL_BWADJ, config.f - 1);
897 writel(reg, priv->base + pll->off);
899 k210_pll_enable(priv, id);
902 return k210_pll_get_rate(priv, id, rate);
905 static ulong k210_pll_set_rate(struct k210_clk_priv *priv, int id, ulong rate,
910 #endif /* CONFIG_CLK_K210_SET_RATE */
912 static ulong k210_pll_get_rate(struct k210_clk_priv *priv, int id,
916 u32 reg = readl(priv->base + k210_plls[id].off);
918 if (reg & K210_PLL_BYPASS)
921 if (!(reg & K210_PLL_PWRD))
924 r = FIELD_GET(K210_PLL_CLKR, reg) + 1;
925 f = FIELD_GET(K210_PLL_CLKF, reg) + 1;
926 od = FIELD_GET(K210_PLL_CLKOD, reg) + 1;
928 return DIV_ROUND_DOWN_ULL(((u64)rate_in) * f, r * od);
932 * Wait for the PLL to be locked. If the PLL is not locked, try clearing the
933 * slip before retrying
935 static void k210_pll_waitfor_lock(struct k210_clk_priv *priv, int id)
937 const struct k210_pll_params *pll = &k210_plls[id];
938 u32 mask = (BIT(pll->width) - 1) << pll->shift;
941 u32 reg = readl(priv->base + K210_SYSCTL_PLL_LOCK);
943 if ((reg & mask) == mask)
946 reg |= BIT(pll->shift + K210_PLL_CLEAR_SLIP);
947 writel(reg, priv->base + K210_SYSCTL_PLL_LOCK);
951 static bool k210_pll_enabled(u32 reg)
953 return (reg & K210_PLL_PWRD) && (reg & K210_PLL_EN) &&
954 !(reg & K210_PLL_RESET);
957 /* Adapted from sysctl_pll_enable */
958 static int k210_pll_enable(struct k210_clk_priv *priv, int id)
960 const struct k210_pll_params *pll = &k210_plls[id];
961 u32 reg = readl(priv->base + pll->off);
963 if (k210_pll_enabled(reg))
966 reg |= K210_PLL_PWRD;
967 writel(reg, priv->base + pll->off);
969 /* Ensure reset is low before asserting it */
970 reg &= ~K210_PLL_RESET;
971 writel(reg, priv->base + pll->off);
972 reg |= K210_PLL_RESET;
973 writel(reg, priv->base + pll->off);
976 reg &= ~K210_PLL_RESET;
977 writel(reg, priv->base + pll->off);
979 k210_pll_waitfor_lock(priv, id);
981 reg &= ~K210_PLL_BYPASS;
983 writel(reg, priv->base + pll->off);
988 static int k210_pll_disable(struct k210_clk_priv *priv, int id)
990 const struct k210_pll_params *pll = &k210_plls[id];
991 u32 reg = readl(priv->base + pll->off);
994 * Bypassing before powering off is important so child clocks don't stop
995 * working. This is especially important for pll0, the indirect parent
998 reg |= K210_PLL_BYPASS;
999 writel(reg, priv->base + pll->off);
1001 reg &= ~K210_PLL_PWRD;
1002 reg &= ~K210_PLL_EN;
1003 writel(reg, priv->base + pll->off);
1007 static u32 k210_clk_readl(struct k210_clk_priv *priv, u8 off, u8 shift,
1010 u32 reg = readl(priv->base + off);
1012 return (reg >> shift) & (BIT(width) - 1);
1015 static void k210_clk_writel(struct k210_clk_priv *priv, u8 off, u8 shift,
1018 u32 reg = readl(priv->base + off);
1019 u32 mask = (BIT(width) - 1) << shift;
1022 reg |= mask & (val << shift);
1023 writel(reg, priv->base + off);
1026 static int k210_clk_get_parent(struct k210_clk_priv *priv, int id)
1029 const struct k210_mux_params *mux;
1031 if (!(k210_clks[id].flags & K210_CLKF_MUX))
1032 return k210_clks[id].parent;
1033 mux = &k210_muxes[k210_clks[id].mux];
1035 sel = k210_clk_readl(priv, mux->off, mux->shift, mux->width);
1036 assert(sel < mux->num_parents);
1037 return mux->parents[sel];
1040 static ulong do_k210_clk_get_rate(struct k210_clk_priv *priv, int id)
1045 const struct k210_div_params *div;
1047 if (id == K210_CLK_IN0)
1048 return clk_get_rate(&priv->in0);
1050 parent = k210_clk_get_parent(priv, id);
1051 parent_rate = do_k210_clk_get_rate(priv, parent);
1052 if (IS_ERR_VALUE(parent_rate))
1055 if (k210_clks[id].flags & K210_CLKF_PLL)
1056 return k210_pll_get_rate(priv, k210_clks[id].pll, parent_rate);
1058 if (k210_clks[id].div == K210_CLK_DIV_NONE)
1060 div = &k210_divs[k210_clks[id].div];
1062 if (div->type == K210_DIV_FIXED)
1063 return parent_rate / div->div;
1065 val = k210_clk_readl(priv, div->off, div->shift, div->width);
1066 switch (div->type) {
1068 return parent_rate / (val + 1);
1070 return parent_rate / 2 / (val + 1);
1071 case K210_DIV_POWER:
1072 /* This is ACLK, which has no divider on IN0 */
1073 if (parent == K210_CLK_IN0)
1075 return parent_rate / (2 << val);
1082 static ulong k210_clk_get_rate(struct clk *clk)
1084 return do_k210_clk_get_rate(dev_get_priv(clk->dev), clk->id);
1087 static int do_k210_clk_set_parent(struct k210_clk_priv *priv, int id, int new)
1090 const struct k210_mux_params *mux;
1092 if (!(k210_clks[id].flags & K210_CLKF_MUX))
1094 mux = &k210_muxes[k210_clks[id].mux];
1096 for (i = 0; i < mux->num_parents; i++) {
1097 if (mux->parents[i] == new) {
1098 k210_clk_writel(priv, mux->off, mux->shift, mux->width,
1106 static int k210_clk_set_parent(struct clk *clk, struct clk *parent)
1108 return do_k210_clk_set_parent(dev_get_priv(clk->dev), clk->id,
1112 static ulong k210_clk_set_rate(struct clk *clk, unsigned long rate)
1114 int parent, ret, err;
1116 const struct k210_div_params *div;
1117 struct k210_clk_priv *priv = dev_get_priv(clk->dev);
1119 if (clk->id == K210_CLK_IN0)
1120 return clk_set_rate(&priv->in0, rate);
1122 parent = k210_clk_get_parent(priv, clk->id);
1123 rate_in = do_k210_clk_get_rate(priv, parent);
1124 if (IS_ERR_VALUE(rate_in))
1127 log_debug("id=%ld rate=%lu rate_in=%lu\n", clk->id, rate, rate_in);
1129 if (clk->id == K210_CLK_PLL0) {
1130 /* Bypass ACLK so the CPU keeps going */
1131 ret = do_k210_clk_set_parent(priv, K210_CLK_ACLK, K210_CLK_IN0);
1134 } else if (clk->id == K210_CLK_PLL1 && gd->flags & GD_FLG_RELOC) {
1136 * We can't bypass the AI clock like we can ACLK, and after
1137 * relocation we are using the AI ram.
1142 if (k210_clks[clk->id].flags & K210_CLKF_PLL) {
1143 ret = k210_pll_set_rate(priv, k210_clks[clk->id].pll, rate,
1145 if (!IS_ERR_VALUE(ret) && clk->id == K210_CLK_PLL0) {
1147 * This may have the side effect of reparenting ACLK,
1148 * but I don't really want to keep track of what the old
1151 err = do_k210_clk_set_parent(priv, K210_CLK_ACLK,
1159 if (k210_clks[clk->id].div == K210_CLK_DIV_NONE)
1161 div = &k210_divs[k210_clks[clk->id].div];
1163 switch (div->type) {
1165 val = DIV_ROUND_CLOSEST_ULL((u64)rate_in, rate);
1166 val = val ? val - 1 : 0;
1169 val = DIV_ROUND_CLOSEST_ULL((u64)rate_in, 2 * rate);
1171 case K210_DIV_POWER:
1172 /* This is ACLK, which has no divider on IN0 */
1173 if (parent == K210_CLK_IN0)
1176 val = DIV_ROUND_CLOSEST_ULL((u64)rate_in, rate);
1184 val = val ? val - 1 : 0;
1185 k210_clk_writel(priv, div->off, div->shift, div->width, val);
1186 return do_k210_clk_get_rate(priv, clk->id);
1189 static int k210_clk_endisable(struct k210_clk_priv *priv, int id, bool enable)
1191 int parent = k210_clk_get_parent(priv, id);
1192 const struct k210_gate_params *gate;
1194 if (id == K210_CLK_IN0) {
1196 return clk_enable(&priv->in0);
1198 return clk_disable(&priv->in0);
1201 /* Only recursively enable clocks since we don't track refcounts */
1203 int ret = k210_clk_endisable(priv, parent, true);
1205 if (ret && ret != -ENOSYS)
1209 if (k210_clks[id].flags & K210_CLKF_PLL) {
1211 return k210_pll_enable(priv, k210_clks[id].pll);
1213 return k210_pll_disable(priv, k210_clks[id].pll);
1216 if (k210_clks[id].gate == K210_CLK_GATE_NONE)
1218 gate = &k210_gates[k210_clks[id].gate];
1220 k210_clk_writel(priv, gate->off, gate->bit_idx, 1, enable);
1224 static int k210_clk_enable(struct clk *clk)
1226 return k210_clk_endisable(dev_get_priv(clk->dev), clk->id, true);
1229 static int k210_clk_disable(struct clk *clk)
1231 return k210_clk_endisable(dev_get_priv(clk->dev), clk->id, false);
1234 static int k210_clk_request(struct clk *clk)
1236 if (clk->id >= ARRAY_SIZE(k210_clks))
1241 static const struct clk_ops k210_clk_ops = {
1242 .request = k210_clk_request,
1243 .set_rate = k210_clk_set_rate,
1244 .get_rate = k210_clk_get_rate,
1245 .set_parent = k210_clk_set_parent,
1246 .enable = k210_clk_enable,
1247 .disable = k210_clk_disable,
1250 static int k210_clk_probe(struct udevice *dev)
1253 struct k210_clk_priv *priv = dev_get_priv(dev);
1255 priv->base = dev_read_addr_ptr(dev_get_parent(dev));
1259 ret = clk_get_by_index(dev, 0, &priv->in0);
1264 * Force setting defaults, even before relocation. This is so we can
1265 * set the clock rate for PLL1 before we relocate into aisram.
1267 if (!(gd->flags & GD_FLG_RELOC))
1268 clk_set_defaults(dev, CLK_DEFAULTS_POST_FORCE);
1273 static const struct udevice_id k210_clk_ids[] = {
1274 { .compatible = "canaan,k210-clk" },
1278 U_BOOT_DRIVER(k210_clk) = {
1281 .of_match = k210_clk_ids,
1282 .ops = &k210_clk_ops,
1283 .probe = k210_clk_probe,
1284 .priv_auto = sizeof(struct k210_clk_priv),
1287 #if IS_ENABLED(CONFIG_CMD_CLK)
1288 static char show_enabled(struct k210_clk_priv *priv, int id)
1292 if (k210_clks[id].flags & K210_CLKF_PLL) {
1293 const struct k210_pll_params *pll =
1294 &k210_plls[k210_clks[id].pll];
1296 enabled = k210_pll_enabled(readl(priv->base + pll->off));
1297 } else if (k210_clks[id].gate == K210_CLK_GATE_NONE) {
1300 const struct k210_gate_params *gate =
1301 &k210_gates[k210_clks[id].gate];
1303 enabled = k210_clk_readl(priv, gate->off, gate->bit_idx, 1);
1306 return enabled ? 'y' : 'n';
1309 static void show_clks(struct k210_clk_priv *priv, int id, int depth)
1313 for (i = 0; i < ARRAY_SIZE(k210_clks); i++) {
1314 if (k210_clk_get_parent(priv, i) != id)
1317 printf(" %-9lu %-7c %*s%s\n", do_k210_clk_get_rate(priv, i),
1318 show_enabled(priv, i), depth * 4, "",
1321 show_clks(priv, i, depth + 1);
1325 int soc_clk_dump(void)
1328 struct udevice *dev;
1329 struct k210_clk_priv *priv;
1331 ret = uclass_get_device_by_driver(UCLASS_CLK, DM_DRIVER_GET(k210_clk),
1335 priv = dev_get_priv(dev);
1337 puts(" Rate Enabled Name\n");
1338 puts("------------------------\n");
1339 printf(" %-9lu %-7c %*s%s\n", clk_get_rate(&priv->in0), 'y', 0, "",
1340 priv->in0.dev->name);
1341 show_clks(priv, K210_CLK_IN0, 1);