1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * drivers/clk/tegra/clk-emc.c
4  *
5  * Copyright (c) 2014, NVIDIA CORPORATION.  All rights reserved.
6  *
7  * Author:
8  *	Mikko Perttunen <mperttunen@nvidia.com>
9  */
10 
11 #include <linux/clk-provider.h>
12 #include <linux/clk.h>
13 #include <linux/clkdev.h>
14 #include <linux/clk/tegra.h>
15 #include <linux/delay.h>
16 #include <linux/export.h>
17 #include <linux/io.h>
18 #include <linux/module.h>
19 #include <linux/of_address.h>
20 #include <linux/of_platform.h>
21 #include <linux/platform_device.h>
22 #include <linux/sort.h>
23 #include <linux/string.h>
24 
25 #include <soc/tegra/fuse.h>
26 
27 #include "clk.h"
28 
29 #define CLK_SOURCE_EMC 0x19c
30 
31 #define CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR_SHIFT 0
32 #define CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR_MASK 0xff
33 #define CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR(x) (((x) & CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR_MASK) << \
34 					      CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR_SHIFT)
35 
36 #define CLK_SOURCE_EMC_EMC_2X_CLK_SRC_SHIFT 29
37 #define CLK_SOURCE_EMC_EMC_2X_CLK_SRC_MASK 0x7
38 #define CLK_SOURCE_EMC_EMC_2X_CLK_SRC(x) (((x) & CLK_SOURCE_EMC_EMC_2X_CLK_SRC_MASK) << \
39 					  CLK_SOURCE_EMC_EMC_2X_CLK_SRC_SHIFT)
40 
41 static const char * const emc_parent_clk_names[] = {
42 	"pll_m", "pll_c", "pll_p", "clk_m", "pll_m_ud",
43 	"pll_c2", "pll_c3", "pll_c_ud"
44 };
45 
46 /*
47  * List of clock sources for various parents the EMC clock can have.
48  * When we change the timing to a timing with a parent that has the same
49  * clock source as the current parent, we must first change to a backup
50  * timing that has a different clock source.
51  */
52 
53 #define EMC_SRC_PLL_M 0
54 #define EMC_SRC_PLL_C 1
55 #define EMC_SRC_PLL_P 2
56 #define EMC_SRC_CLK_M 3
57 #define EMC_SRC_PLL_C2 4
58 #define EMC_SRC_PLL_C3 5
59 
60 static const char emc_parent_clk_sources[] = {
61 	EMC_SRC_PLL_M, EMC_SRC_PLL_C, EMC_SRC_PLL_P, EMC_SRC_CLK_M,
62 	EMC_SRC_PLL_M, EMC_SRC_PLL_C2, EMC_SRC_PLL_C3, EMC_SRC_PLL_C
63 };
64 
65 struct emc_timing {
66 	unsigned long rate, parent_rate;
67 	u8 parent_index;
68 	struct clk *parent;
69 	u32 ram_code;
70 };
71 
72 struct tegra_clk_emc {
73 	struct clk_hw hw;
74 	void __iomem *clk_regs;
75 	struct clk *prev_parent;
76 	bool changing_timing;
77 
78 	struct device_node *emc_node;
79 	struct tegra_emc *emc;
80 
81 	int num_timings;
82 	struct emc_timing *timings;
83 	spinlock_t *lock;
84 
85 	tegra124_emc_prepare_timing_change_cb *prepare_timing_change;
86 	tegra124_emc_complete_timing_change_cb *complete_timing_change;
87 };
88 
89 /* Common clock framework callback implementations */
90 
91 static unsigned long emc_recalc_rate(struct clk_hw *hw,
92 				     unsigned long parent_rate)
93 {
94 	struct tegra_clk_emc *tegra;
95 	u32 val, div;
96 
97 	tegra = container_of(hw, struct tegra_clk_emc, hw);
98 
99 	/*
100 	 * CCF wrongly assumes that the parent won't change during set_rate,
101 	 * so get the parent rate explicitly.
102 	 */
103 	parent_rate = clk_hw_get_rate(clk_hw_get_parent(hw));
104 
105 	val = readl(tegra->clk_regs + CLK_SOURCE_EMC);
106 	div = val & CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR_MASK;
107 
108 	return parent_rate / (div + 2) * 2;
109 }
110 
111 /*
112  * Rounds up unless no higher rate exists, in which case down. This way is
113  * safer since things have EMC rate floors. Also don't touch parent_rate
114  * since we don't want the CCF to play with our parent clocks.
115  */
116 static int emc_determine_rate(struct clk_hw *hw, struct clk_rate_request *req)
117 {
118 	struct tegra_clk_emc *tegra;
119 	u8 ram_code = tegra_read_ram_code();
120 	struct emc_timing *timing = NULL;
121 	int i, k, t;
122 
123 	tegra = container_of(hw, struct tegra_clk_emc, hw);
124 
125 	for (k = 0; k < tegra->num_timings; k++) {
126 		if (tegra->timings[k].ram_code == ram_code)
127 			break;
128 	}
129 
130 	for (t = k; t < tegra->num_timings; t++) {
131 		if (tegra->timings[t].ram_code != ram_code)
132 			break;
133 	}
134 
135 	for (i = k; i < t; i++) {
136 		timing = tegra->timings + i;
137 
138 		if (timing->rate < req->rate && i != t - 1)
139 			continue;
140 
141 		if (timing->rate > req->max_rate) {
142 			i = max(i, k + 1);
143 			req->rate = tegra->timings[i - 1].rate;
144 			return 0;
145 		}
146 
147 		if (timing->rate < req->min_rate)
148 			continue;
149 
150 		req->rate = timing->rate;
151 		return 0;
152 	}
153 
154 	if (timing) {
155 		req->rate = timing->rate;
156 		return 0;
157 	}
158 
159 	req->rate = clk_hw_get_rate(hw);
160 	return 0;
161 }
162 
163 static u8 emc_get_parent(struct clk_hw *hw)
164 {
165 	struct tegra_clk_emc *tegra;
166 	u32 val;
167 
168 	tegra = container_of(hw, struct tegra_clk_emc, hw);
169 
170 	val = readl(tegra->clk_regs + CLK_SOURCE_EMC);
171 
172 	return (val >> CLK_SOURCE_EMC_EMC_2X_CLK_SRC_SHIFT)
173 		& CLK_SOURCE_EMC_EMC_2X_CLK_SRC_MASK;
174 }
175 
176 static struct tegra_emc *emc_ensure_emc_driver(struct tegra_clk_emc *tegra)
177 {
178 	struct platform_device *pdev;
179 
180 	if (tegra->emc)
181 		return tegra->emc;
182 
183 	if (!tegra->prepare_timing_change || !tegra->complete_timing_change)
184 		return NULL;
185 
186 	if (!tegra->emc_node)
187 		return NULL;
188 
189 	pdev = of_find_device_by_node(tegra->emc_node);
190 	if (!pdev) {
191 		pr_err("%s: could not get external memory controller\n",
192 		       __func__);
193 		return NULL;
194 	}
195 
196 	of_node_put(tegra->emc_node);
197 	tegra->emc_node = NULL;
198 
199 	tegra->emc = platform_get_drvdata(pdev);
200 	if (!tegra->emc) {
201 		pr_err("%s: cannot find EMC driver\n", __func__);
202 		return NULL;
203 	}
204 
205 	return tegra->emc;
206 }
207 
208 static int emc_set_timing(struct tegra_clk_emc *tegra,
209 			  struct emc_timing *timing)
210 {
211 	int err;
212 	u8 div;
213 	u32 car_value;
214 	unsigned long flags = 0;
215 	struct tegra_emc *emc = emc_ensure_emc_driver(tegra);
216 
217 	if (!emc)
218 		return -ENOENT;
219 
220 	pr_debug("going to rate %ld prate %ld p %s\n", timing->rate,
221 		 timing->parent_rate, __clk_get_name(timing->parent));
222 
223 	if (emc_get_parent(&tegra->hw) == timing->parent_index &&
224 	    clk_get_rate(timing->parent) != timing->parent_rate) {
225 		WARN_ONCE(1, "parent %s rate mismatch %lu %lu\n",
226 			  __clk_get_name(timing->parent),
227 			  clk_get_rate(timing->parent),
228 			  timing->parent_rate);
229 		return -EINVAL;
230 	}
231 
232 	tegra->changing_timing = true;
233 
234 	err = clk_set_rate(timing->parent, timing->parent_rate);
235 	if (err) {
236 		pr_err("cannot change parent %s rate to %ld: %d\n",
237 		       __clk_get_name(timing->parent), timing->parent_rate,
238 		       err);
239 
240 		return err;
241 	}
242 
243 	err = clk_prepare_enable(timing->parent);
244 	if (err) {
245 		pr_err("cannot enable parent clock: %d\n", err);
246 		return err;
247 	}
248 
249 	div = timing->parent_rate / (timing->rate / 2) - 2;
250 
251 	err = tegra->prepare_timing_change(emc, timing->rate);
252 	if (err)
253 		return err;
254 
255 	spin_lock_irqsave(tegra->lock, flags);
256 
257 	car_value = readl(tegra->clk_regs + CLK_SOURCE_EMC);
258 
259 	car_value &= ~CLK_SOURCE_EMC_EMC_2X_CLK_SRC(~0);
260 	car_value |= CLK_SOURCE_EMC_EMC_2X_CLK_SRC(timing->parent_index);
261 
262 	car_value &= ~CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR(~0);
263 	car_value |= CLK_SOURCE_EMC_EMC_2X_CLK_DIVISOR(div);
264 
265 	writel(car_value, tegra->clk_regs + CLK_SOURCE_EMC);
266 
267 	spin_unlock_irqrestore(tegra->lock, flags);
268 
269 	tegra->complete_timing_change(emc, timing->rate);
270 
271 	clk_hw_reparent(&tegra->hw, __clk_get_hw(timing->parent));
272 	clk_disable_unprepare(tegra->prev_parent);
273 
274 	tegra->prev_parent = timing->parent;
275 	tegra->changing_timing = false;
276 
277 	return 0;
278 }
279 
280 /*
281  * Get backup timing to use as an intermediate step when a change between
282  * two timings with the same clock source has been requested. First try to
283  * find a timing with a higher clock rate to avoid a rate below any set rate
284  * floors. If that is not possible, find a lower rate.
285  */
286 static struct emc_timing *get_backup_timing(struct tegra_clk_emc *tegra,
287 					    int timing_index)
288 {
289 	int i;
290 	u32 ram_code = tegra_read_ram_code();
291 	struct emc_timing *timing;
292 
293 	for (i = timing_index+1; i < tegra->num_timings; i++) {
294 		timing = tegra->timings + i;
295 		if (timing->ram_code != ram_code)
296 			break;
297 
298 		if (emc_parent_clk_sources[timing->parent_index] !=
299 		    emc_parent_clk_sources[
300 		      tegra->timings[timing_index].parent_index])
301 			return timing;
302 	}
303 
304 	for (i = timing_index-1; i >= 0; --i) {
305 		timing = tegra->timings + i;
306 		if (timing->ram_code != ram_code)
307 			break;
308 
309 		if (emc_parent_clk_sources[timing->parent_index] !=
310 		    emc_parent_clk_sources[
311 		      tegra->timings[timing_index].parent_index])
312 			return timing;
313 	}
314 
315 	return NULL;
316 }
317 
318 static int emc_set_rate(struct clk_hw *hw, unsigned long rate,
319 			unsigned long parent_rate)
320 {
321 	struct tegra_clk_emc *tegra;
322 	struct emc_timing *timing = NULL;
323 	int i, err;
324 	u32 ram_code = tegra_read_ram_code();
325 
326 	tegra = container_of(hw, struct tegra_clk_emc, hw);
327 
328 	if (clk_hw_get_rate(hw) == rate)
329 		return 0;
330 
331 	/*
332 	 * When emc_set_timing changes the parent rate, CCF will propagate
333 	 * that downward to us, so ignore any set_rate calls while a rate
334 	 * change is already going on.
335 	 */
336 	if (tegra->changing_timing)
337 		return 0;
338 
339 	for (i = 0; i < tegra->num_timings; i++) {
340 		if (tegra->timings[i].rate == rate &&
341 		    tegra->timings[i].ram_code == ram_code) {
342 			timing = tegra->timings + i;
343 			break;
344 		}
345 	}
346 
347 	if (!timing) {
348 		pr_err("cannot switch to rate %ld without emc table\n", rate);
349 		return -EINVAL;
350 	}
351 
352 	if (emc_parent_clk_sources[emc_get_parent(hw)] ==
353 	    emc_parent_clk_sources[timing->parent_index] &&
354 	    clk_get_rate(timing->parent) != timing->parent_rate) {
355 		/*
356 		 * Parent clock source not changed but parent rate has changed,
357 		 * need to temporarily switch to another parent
358 		 */
359 
360 		struct emc_timing *backup_timing;
361 
362 		backup_timing = get_backup_timing(tegra, i);
363 		if (!backup_timing) {
364 			pr_err("cannot find backup timing\n");
365 			return -EINVAL;
366 		}
367 
368 		pr_debug("using %ld as backup rate when going to %ld\n",
369 			 backup_timing->rate, rate);
370 
371 		err = emc_set_timing(tegra, backup_timing);
372 		if (err) {
373 			pr_err("cannot set backup timing: %d\n", err);
374 			return err;
375 		}
376 	}
377 
378 	return emc_set_timing(tegra, timing);
379 }
380 
381 /* Initialization and deinitialization */
382 
383 static int load_one_timing_from_dt(struct tegra_clk_emc *tegra,
384 				   struct emc_timing *timing,
385 				   struct device_node *node)
386 {
387 	int err, i;
388 	u32 tmp;
389 
390 	err = of_property_read_u32(node, "clock-frequency", &tmp);
391 	if (err) {
392 		pr_err("timing %pOF: failed to read rate\n", node);
393 		return err;
394 	}
395 
396 	timing->rate = tmp;
397 
398 	err = of_property_read_u32(node, "nvidia,parent-clock-frequency", &tmp);
399 	if (err) {
400 		pr_err("timing %pOF: failed to read parent rate\n", node);
401 		return err;
402 	}
403 
404 	timing->parent_rate = tmp;
405 
406 	timing->parent = of_clk_get_by_name(node, "emc-parent");
407 	if (IS_ERR(timing->parent)) {
408 		pr_err("timing %pOF: failed to get parent clock\n", node);
409 		return PTR_ERR(timing->parent);
410 	}
411 
412 	timing->parent_index = 0xff;
413 	i = match_string(emc_parent_clk_names, ARRAY_SIZE(emc_parent_clk_names),
414 			 __clk_get_name(timing->parent));
415 	if (i < 0) {
416 		pr_err("timing %pOF: %s is not a valid parent\n",
417 		       node, __clk_get_name(timing->parent));
418 		clk_put(timing->parent);
419 		return -EINVAL;
420 	}
421 
422 	timing->parent_index = i;
423 	return 0;
424 }
425 
426 static int cmp_timings(const void *_a, const void *_b)
427 {
428 	const struct emc_timing *a = _a;
429 	const struct emc_timing *b = _b;
430 
431 	if (a->rate < b->rate)
432 		return -1;
433 	else if (a->rate == b->rate)
434 		return 0;
435 	else
436 		return 1;
437 }
438 
439 static int load_timings_from_dt(struct tegra_clk_emc *tegra,
440 				struct device_node *node,
441 				u32 ram_code)
442 {
443 	struct emc_timing *timings_ptr;
444 	struct device_node *child;
445 	int child_count = of_get_child_count(node);
446 	int i = 0, err;
447 	size_t size;
448 
449 	size = (tegra->num_timings + child_count) * sizeof(struct emc_timing);
450 
451 	tegra->timings = krealloc(tegra->timings, size, GFP_KERNEL);
452 	if (!tegra->timings)
453 		return -ENOMEM;
454 
455 	timings_ptr = tegra->timings + tegra->num_timings;
456 	tegra->num_timings += child_count;
457 
458 	for_each_child_of_node(node, child) {
459 		struct emc_timing *timing = timings_ptr + (i++);
460 
461 		err = load_one_timing_from_dt(tegra, timing, child);
462 		if (err) {
463 			of_node_put(child);
464 			return err;
465 		}
466 
467 		timing->ram_code = ram_code;
468 	}
469 
470 	sort(timings_ptr, child_count, sizeof(struct emc_timing),
471 	     cmp_timings, NULL);
472 
473 	return 0;
474 }
475 
476 static const struct clk_ops tegra_clk_emc_ops = {
477 	.recalc_rate = emc_recalc_rate,
478 	.determine_rate = emc_determine_rate,
479 	.set_rate = emc_set_rate,
480 	.get_parent = emc_get_parent,
481 };
482 
483 struct clk *tegra124_clk_register_emc(void __iomem *base, struct device_node *np,
484 				      spinlock_t *lock)
485 {
486 	struct tegra_clk_emc *tegra;
487 	struct clk_init_data init;
488 	struct device_node *node;
489 	u32 node_ram_code;
490 	struct clk *clk;
491 	int err;
492 
493 	tegra = kcalloc(1, sizeof(*tegra), GFP_KERNEL);
494 	if (!tegra)
495 		return ERR_PTR(-ENOMEM);
496 
497 	tegra->clk_regs = base;
498 	tegra->lock = lock;
499 
500 	tegra->num_timings = 0;
501 
502 	for_each_child_of_node(np, node) {
503 		err = of_property_read_u32(node, "nvidia,ram-code",
504 					   &node_ram_code);
505 		if (err)
506 			continue;
507 
508 		/*
509 		 * Store timings for all ram codes as we cannot read the
510 		 * fuses until the apbmisc driver is loaded.
511 		 */
512 		err = load_timings_from_dt(tegra, node, node_ram_code);
513 		if (err) {
514 			of_node_put(node);
515 			return ERR_PTR(err);
516 		}
517 	}
518 
519 	if (tegra->num_timings == 0)
520 		pr_warn("%s: no memory timings registered\n", __func__);
521 
522 	tegra->emc_node = of_parse_phandle(np,
523 			"nvidia,external-memory-controller", 0);
524 	if (!tegra->emc_node)
525 		pr_warn("%s: couldn't find node for EMC driver\n", __func__);
526 
527 	init.name = "emc";
528 	init.ops = &tegra_clk_emc_ops;
529 	init.flags = CLK_IS_CRITICAL;
530 	init.parent_names = emc_parent_clk_names;
531 	init.num_parents = ARRAY_SIZE(emc_parent_clk_names);
532 
533 	tegra->hw.init = &init;
534 
535 	clk = clk_register(NULL, &tegra->hw);
536 	if (IS_ERR(clk))
537 		return clk;
538 
539 	tegra->prev_parent = clk_hw_get_parent_by_index(
540 		&tegra->hw, emc_get_parent(&tegra->hw))->clk;
541 	tegra->changing_timing = false;
542 
543 	/* Allow debugging tools to see the EMC clock */
544 	clk_register_clkdev(clk, "emc", "tegra-clk-debug");
545 
546 	return clk;
547 };
548 
549 void tegra124_clk_set_emc_callbacks(tegra124_emc_prepare_timing_change_cb *prep_cb,
550 				    tegra124_emc_complete_timing_change_cb *complete_cb)
551 {
552 	struct clk *clk = __clk_lookup("emc");
553 	struct tegra_clk_emc *tegra;
554 	struct clk_hw *hw;
555 
556 	if (clk) {
557 		hw = __clk_get_hw(clk);
558 		tegra = container_of(hw, struct tegra_clk_emc, hw);
559 
560 		tegra->prepare_timing_change = prep_cb;
561 		tegra->complete_timing_change = complete_cb;
562 	}
563 }
564 EXPORT_SYMBOL_GPL(tegra124_clk_set_emc_callbacks);
565 
566 bool tegra124_clk_emc_driver_available(struct clk_hw *hw)
567 {
568 	struct tegra_clk_emc *tegra = container_of(hw, struct tegra_clk_emc, hw);
569 
570 	return tegra->prepare_timing_change && tegra->complete_timing_change;
571 }
572