1/*
  2 * ARM big.LITTLE Platforms CPUFreq support
  3 *
  4 * Copyright (C) 2013 ARM Ltd.
  5 * Sudeep KarkadaNagesha <sudeep.karkadanagesha@arm.com>
  6 *
  7 * Copyright (C) 2013 Linaro.
  8 * Viresh Kumar <viresh.kumar@linaro.org>
  9 *
 10 * This program is free software; you can redistribute it and/or modify
 11 * it under the terms of the GNU General Public License version 2 as
 12 * published by the Free Software Foundation.
 13 *
 14 * This program is distributed "as is" WITHOUT ANY WARRANTY of any
 15 * kind, whether express or implied; without even the implied warranty
 16 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 17 * GNU General Public License for more details.
 18 */
 19
 20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 21
 22#include <linux/clk.h>
 23#include <linux/cpu.h>
 24#include <linux/cpufreq.h>
 25#include <linux/cpumask.h>
 26#include <linux/cpu_cooling.h>
 27#include <linux/export.h>
 28#include <linux/module.h>
 29#include <linux/mutex.h>
 30#include <linux/of_platform.h>
 31#include <linux/pm_opp.h>
 32#include <linux/slab.h>
 33#include <linux/topology.h>
 34#include <linux/types.h>
 35
 36#include "arm_big_little.h"
 37
 38/* Currently we support only two clusters */
 39#define A15_CLUSTER	0
 40#define A7_CLUSTER	1
 41#define MAX_CLUSTERS	2
 42
 43#ifdef CONFIG_BL_SWITCHER
 44#include <asm/bL_switcher.h>
 45static bool bL_switching_enabled;
 46#define is_bL_switching_enabled()	bL_switching_enabled
 47#define set_switching_enabled(x)	(bL_switching_enabled = (x))
 48#else
 49#define is_bL_switching_enabled()	false
 50#define set_switching_enabled(x)	do { } while (0)
 51#define bL_switch_request(...)		do { } while (0)
 52#define bL_switcher_put_enabled()	do { } while (0)
 53#define bL_switcher_get_enabled()	do { } while (0)
 54#endif
 55
 56#define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
 57#define VIRT_FREQ(cluster, freq)    ((cluster == A7_CLUSTER) ? freq >> 1 : freq)
 58
 59static struct thermal_cooling_device *cdev[MAX_CLUSTERS];
 60static struct cpufreq_arm_bL_ops *arm_bL_ops;
 61static struct clk *clk[MAX_CLUSTERS];
 62static struct cpufreq_frequency_table *freq_table[MAX_CLUSTERS + 1];
 63static atomic_t cluster_usage[MAX_CLUSTERS + 1];
 64
 65static unsigned int clk_big_min;	/* (Big) clock frequencies */
 66static unsigned int clk_little_max;	/* Maximum clock frequency (Little) */
 67
 68static DEFINE_PER_CPU(unsigned int, physical_cluster);
 69static DEFINE_PER_CPU(unsigned int, cpu_last_req_freq);
 70
 71static struct mutex cluster_lock[MAX_CLUSTERS];
 72
 73static inline int raw_cpu_to_cluster(int cpu)
 74{
 75	return topology_physical_package_id(cpu);
 76}
 77
 78static inline int cpu_to_cluster(int cpu)
 79{
 80	return is_bL_switching_enabled() ?
 81		MAX_CLUSTERS : raw_cpu_to_cluster(cpu);
 82}
 83
 84static unsigned int find_cluster_maxfreq(int cluster)
 85{
 86	int j;
 87	u32 max_freq = 0, cpu_freq;
 88
 89	for_each_online_cpu(j) {
 90		cpu_freq = per_cpu(cpu_last_req_freq, j);
 91
 92		if ((cluster == per_cpu(physical_cluster, j)) &&
 93				(max_freq < cpu_freq))
 94			max_freq = cpu_freq;
 95	}
 96
 97	pr_debug("%s: cluster: %d, max freq: %d\n", __func__, cluster,
 98			max_freq);
 99
100	return max_freq;
101}
102
103static unsigned int clk_get_cpu_rate(unsigned int cpu)
104{
105	u32 cur_cluster = per_cpu(physical_cluster, cpu);
106	u32 rate = clk_get_rate(clk[cur_cluster]) / 1000;
107
108	/* For switcher we use virtual A7 clock rates */
109	if (is_bL_switching_enabled())
110		rate = VIRT_FREQ(cur_cluster, rate);
111
112	pr_debug("%s: cpu: %d, cluster: %d, freq: %u\n", __func__, cpu,
113			cur_cluster, rate);
114
115	return rate;
116}
117
118static unsigned int bL_cpufreq_get_rate(unsigned int cpu)
119{
120	if (is_bL_switching_enabled()) {
121		pr_debug("%s: freq: %d\n", __func__, per_cpu(cpu_last_req_freq,
122					cpu));
123
124		return per_cpu(cpu_last_req_freq, cpu);
125	} else {
126		return clk_get_cpu_rate(cpu);
127	}
128}
129
130static unsigned int
131bL_cpufreq_set_rate(u32 cpu, u32 old_cluster, u32 new_cluster, u32 rate)
132{
133	u32 new_rate, prev_rate;
134	int ret;
135	bool bLs = is_bL_switching_enabled();
136
137	mutex_lock(&cluster_lock[new_cluster]);
138
139	if (bLs) {
140		prev_rate = per_cpu(cpu_last_req_freq, cpu);
141		per_cpu(cpu_last_req_freq, cpu) = rate;
142		per_cpu(physical_cluster, cpu) = new_cluster;
143
144		new_rate = find_cluster_maxfreq(new_cluster);
145		new_rate = ACTUAL_FREQ(new_cluster, new_rate);
146	} else {
147		new_rate = rate;
148	}
149
150	pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d, freq: %d\n",
151			__func__, cpu, old_cluster, new_cluster, new_rate);
152
153	ret = clk_set_rate(clk[new_cluster], new_rate * 1000);
154	if (!ret) {
155		/*
156		 * FIXME: clk_set_rate hasn't returned an error here however it
157		 * may be that clk_change_rate failed due to hardware or
158		 * firmware issues and wasn't able to report that due to the
159		 * current design of the clk core layer. To work around this
160		 * problem we will read back the clock rate and check it is
161		 * correct. This needs to be removed once clk core is fixed.
162		 */
163		if (clk_get_rate(clk[new_cluster]) != new_rate * 1000)
164			ret = -EIO;
165	}
166
167	if (WARN_ON(ret)) {
168		pr_err("clk_set_rate failed: %d, new cluster: %d\n", ret,
169				new_cluster);
170		if (bLs) {
171			per_cpu(cpu_last_req_freq, cpu) = prev_rate;
172			per_cpu(physical_cluster, cpu) = old_cluster;
173		}
174
175		mutex_unlock(&cluster_lock[new_cluster]);
176
177		return ret;
178	}
179
180	mutex_unlock(&cluster_lock[new_cluster]);
181
182	/* Recalc freq for old cluster when switching clusters */
183	if (old_cluster != new_cluster) {
184		pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d\n",
185				__func__, cpu, old_cluster, new_cluster);
186
187		/* Switch cluster */
188		bL_switch_request(cpu, new_cluster);
189
190		mutex_lock(&cluster_lock[old_cluster]);
191
192		/* Set freq of old cluster if there are cpus left on it */
193		new_rate = find_cluster_maxfreq(old_cluster);
194		new_rate = ACTUAL_FREQ(old_cluster, new_rate);
195
196		if (new_rate) {
197			pr_debug("%s: Updating rate of old cluster: %d, to freq: %d\n",
198					__func__, old_cluster, new_rate);
199
200			if (clk_set_rate(clk[old_cluster], new_rate * 1000))
201				pr_err("%s: clk_set_rate failed: %d, old cluster: %d\n",
202						__func__, ret, old_cluster);
203		}
204		mutex_unlock(&cluster_lock[old_cluster]);
205	}
206
207	return 0;
208}
209
210/* Set clock frequency */
211static int bL_cpufreq_set_target(struct cpufreq_policy *policy,
212		unsigned int index)
213{
214	u32 cpu = policy->cpu, cur_cluster, new_cluster, actual_cluster;
215	unsigned int freqs_new;
216
217	cur_cluster = cpu_to_cluster(cpu);
218	new_cluster = actual_cluster = per_cpu(physical_cluster, cpu);
219
220	freqs_new = freq_table[cur_cluster][index].frequency;
221
222	if (is_bL_switching_enabled()) {
223		if ((actual_cluster == A15_CLUSTER) &&
224				(freqs_new < clk_big_min)) {
225			new_cluster = A7_CLUSTER;
226		} else if ((actual_cluster == A7_CLUSTER) &&
227				(freqs_new > clk_little_max)) {
228			new_cluster = A15_CLUSTER;
229		}
230	}
231
232	return bL_cpufreq_set_rate(cpu, actual_cluster, new_cluster, freqs_new);
233}
234
235static inline u32 get_table_count(struct cpufreq_frequency_table *table)
236{
237	int count;
238
239	for (count = 0; table[count].frequency != CPUFREQ_TABLE_END; count++)
240		;
241
242	return count;
243}
244
245/* get the minimum frequency in the cpufreq_frequency_table */
246static inline u32 get_table_min(struct cpufreq_frequency_table *table)
247{
248	struct cpufreq_frequency_table *pos;
249	uint32_t min_freq = ~0;
250	cpufreq_for_each_entry(pos, table)
251		if (pos->frequency < min_freq)
252			min_freq = pos->frequency;
253	return min_freq;
254}
255
256/* get the maximum frequency in the cpufreq_frequency_table */
257static inline u32 get_table_max(struct cpufreq_frequency_table *table)
258{
259	struct cpufreq_frequency_table *pos;
260	uint32_t max_freq = 0;
261	cpufreq_for_each_entry(pos, table)
262		if (pos->frequency > max_freq)
263			max_freq = pos->frequency;
264	return max_freq;
265}
266
267static int merge_cluster_tables(void)
268{
269	int i, j, k = 0, count = 1;
270	struct cpufreq_frequency_table *table;
271
272	for (i = 0; i < MAX_CLUSTERS; i++)
273		count += get_table_count(freq_table[i]);
274
275	table = kzalloc(sizeof(*table) * count, GFP_KERNEL);
276	if (!table)
277		return -ENOMEM;
278
279	freq_table[MAX_CLUSTERS] = table;
280
281	/* Add in reverse order to get freqs in increasing order */
282	for (i = MAX_CLUSTERS - 1; i >= 0; i--) {
283		for (j = 0; freq_table[i][j].frequency != CPUFREQ_TABLE_END;
284				j++) {
285			table[k].frequency = VIRT_FREQ(i,
286					freq_table[i][j].frequency);
287			pr_debug("%s: index: %d, freq: %d\n", __func__, k,
288					table[k].frequency);
289			k++;
290		}
291	}
292
293	table[k].driver_data = k;
294	table[k].frequency = CPUFREQ_TABLE_END;
295
296	pr_debug("%s: End, table: %p, count: %d\n", __func__, table, k);
297
298	return 0;
299}
300
301static void _put_cluster_clk_and_freq_table(struct device *cpu_dev)
302{
303	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
304
305	if (!freq_table[cluster])
306		return;
307
308	clk_put(clk[cluster]);
309	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
310	if (arm_bL_ops->free_opp_table)
311		arm_bL_ops->free_opp_table(cpu_dev);
312	dev_dbg(cpu_dev, "%s: cluster: %d\n", __func__, cluster);
313}
314
315static void put_cluster_clk_and_freq_table(struct device *cpu_dev)
316{
317	u32 cluster = cpu_to_cluster(cpu_dev->id);
318	int i;
319
320	if (atomic_dec_return(&cluster_usage[cluster]))
321		return;
322
323	if (cluster < MAX_CLUSTERS)
324		return _put_cluster_clk_and_freq_table(cpu_dev);
325
326	for_each_present_cpu(i) {
327		struct device *cdev = get_cpu_device(i);
328		if (!cdev) {
329			pr_err("%s: failed to get cpu%d device\n", __func__, i);
330			return;
331		}
332
333		_put_cluster_clk_and_freq_table(cdev);
334	}
335
336	/* free virtual table */
337	kfree(freq_table[cluster]);
338}
339
340static int _get_cluster_clk_and_freq_table(struct device *cpu_dev)
341{
342	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
343	int ret;
344
345	if (freq_table[cluster])
346		return 0;
347
348	ret = arm_bL_ops->init_opp_table(cpu_dev);
349	if (ret) {
350		dev_err(cpu_dev, "%s: init_opp_table failed, cpu: %d, err: %d\n",
351				__func__, cpu_dev->id, ret);
352		goto out;
353	}
354
355	ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster]);
356	if (ret) {
357		dev_err(cpu_dev, "%s: failed to init cpufreq table, cpu: %d, err: %d\n",
358				__func__, cpu_dev->id, ret);
359		goto free_opp_table;
360	}
361
362	clk[cluster] = clk_get(cpu_dev, NULL);
363	if (!IS_ERR(clk[cluster])) {
364		dev_dbg(cpu_dev, "%s: clk: %p & freq table: %p, cluster: %d\n",
365				__func__, clk[cluster], freq_table[cluster],
366				cluster);
367		return 0;
368	}
369
370	dev_err(cpu_dev, "%s: Failed to get clk for cpu: %d, cluster: %d\n",
371			__func__, cpu_dev->id, cluster);
372	ret = PTR_ERR(clk[cluster]);
373	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
374
375free_opp_table:
376	if (arm_bL_ops->free_opp_table)
377		arm_bL_ops->free_opp_table(cpu_dev);
378out:
379	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n", __func__,
380			cluster);
381	return ret;
382}
383
384static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
385{
386	u32 cluster = cpu_to_cluster(cpu_dev->id);
387	int i, ret;
388
389	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
390		return 0;
391
392	if (cluster < MAX_CLUSTERS) {
393		ret = _get_cluster_clk_and_freq_table(cpu_dev);
394		if (ret)
395			atomic_dec(&cluster_usage[cluster]);
396		return ret;
397	}
398
399	/*
400	 * Get data for all clusters and fill virtual cluster with a merge of
401	 * both
402	 */
403	for_each_present_cpu(i) {
404		struct device *cdev = get_cpu_device(i);
405		if (!cdev) {
406			pr_err("%s: failed to get cpu%d device\n", __func__, i);
407			return -ENODEV;
408		}
409
410		ret = _get_cluster_clk_and_freq_table(cdev);
411		if (ret)
412			goto put_clusters;
413	}
414
415	ret = merge_cluster_tables();
416	if (ret)
417		goto put_clusters;
418
419	/* Assuming 2 cluster, set clk_big_min and clk_little_max */
420	clk_big_min = get_table_min(freq_table[0]);
421	clk_little_max = VIRT_FREQ(1, get_table_max(freq_table[1]));
422
423	pr_debug("%s: cluster: %d, clk_big_min: %d, clk_little_max: %d\n",
424			__func__, cluster, clk_big_min, clk_little_max);
425
426	return 0;
427
428put_clusters:
429	for_each_present_cpu(i) {
430		struct device *cdev = get_cpu_device(i);
431		if (!cdev) {
432			pr_err("%s: failed to get cpu%d device\n", __func__, i);
433			return -ENODEV;
434		}
435
436		_put_cluster_clk_and_freq_table(cdev);
437	}
438
439	atomic_dec(&cluster_usage[cluster]);
440
441	return ret;
442}
443
444/* Per-CPU initialization */
445static int bL_cpufreq_init(struct cpufreq_policy *policy)
446{
447	u32 cur_cluster = cpu_to_cluster(policy->cpu);
448	struct device *cpu_dev;
449	int ret;
450
451	cpu_dev = get_cpu_device(policy->cpu);
452	if (!cpu_dev) {
453		pr_err("%s: failed to get cpu%d device\n", __func__,
454				policy->cpu);
455		return -ENODEV;
456	}
457
458	ret = get_cluster_clk_and_freq_table(cpu_dev);
459	if (ret)
460		return ret;
461
462	ret = cpufreq_table_validate_and_show(policy, freq_table[cur_cluster]);
463	if (ret) {
464		dev_err(cpu_dev, "CPU %d, cluster: %d invalid freq table\n",
465				policy->cpu, cur_cluster);
466		put_cluster_clk_and_freq_table(cpu_dev);
467		return ret;
468	}
469
470	if (cur_cluster < MAX_CLUSTERS) {
471		int cpu;
472
473		cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
474
475		for_each_cpu(cpu, policy->cpus)
476			per_cpu(physical_cluster, cpu) = cur_cluster;
477	} else {
478		/* Assumption: during init, we are always running on A15 */
479		per_cpu(physical_cluster, policy->cpu) = A15_CLUSTER;
480	}
481
482	if (arm_bL_ops->get_transition_latency)
483		policy->cpuinfo.transition_latency =
484			arm_bL_ops->get_transition_latency(cpu_dev);
485	else
486		policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
487
488	if (is_bL_switching_enabled())
489		per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate(policy->cpu);
490
491	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
492	return 0;
493}
494
495static int bL_cpufreq_exit(struct cpufreq_policy *policy)
496{
497	struct device *cpu_dev;
498	int cur_cluster = cpu_to_cluster(policy->cpu);
499
500	if (cur_cluster < MAX_CLUSTERS) {
501		cpufreq_cooling_unregister(cdev[cur_cluster]);
502		cdev[cur_cluster] = NULL;
503	}
504
505	cpu_dev = get_cpu_device(policy->cpu);
506	if (!cpu_dev) {
507		pr_err("%s: failed to get cpu%d device\n", __func__,
508				policy->cpu);
509		return -ENODEV;
510	}
511
512	put_cluster_clk_and_freq_table(cpu_dev);
513	dev_dbg(cpu_dev, "%s: Exited, cpu: %d\n", __func__, policy->cpu);
514
515	return 0;
516}
517
518static void bL_cpufreq_ready(struct cpufreq_policy *policy)
519{
520	struct device *cpu_dev = get_cpu_device(policy->cpu);
521	int cur_cluster = cpu_to_cluster(policy->cpu);
522	struct device_node *np;
523
524	/* Do not register a cpu_cooling device if we are in IKS mode */
525	if (cur_cluster >= MAX_CLUSTERS)
526		return;
527
528	np = of_node_get(cpu_dev->of_node);
529	if (WARN_ON(!np))
530		return;
531
532	if (of_find_property(np, "#cooling-cells", NULL)) {
533		u32 power_coefficient = 0;
534
535		of_property_read_u32(np, "dynamic-power-coefficient",
536				     &power_coefficient);
537
538		cdev[cur_cluster] = of_cpufreq_power_cooling_register(np,
539				policy->related_cpus, power_coefficient, NULL);
540		if (IS_ERR(cdev[cur_cluster])) {
541			dev_err(cpu_dev,
542				"running cpufreq without cooling device: %ld\n",
543				PTR_ERR(cdev[cur_cluster]));
544			cdev[cur_cluster] = NULL;
545		}
546	}
547	of_node_put(np);
548}
549
550static struct cpufreq_driver bL_cpufreq_driver = {
551	.name			= "arm-big-little",
552	.flags			= CPUFREQ_STICKY |
553					CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
554					CPUFREQ_NEED_INITIAL_FREQ_CHECK,
555	.verify			= cpufreq_generic_frequency_table_verify,
556	.target_index		= bL_cpufreq_set_target,
557	.get			= bL_cpufreq_get_rate,
558	.init			= bL_cpufreq_init,
559	.exit			= bL_cpufreq_exit,
560	.ready			= bL_cpufreq_ready,
561	.attr			= cpufreq_generic_attr,
562};
563
564#ifdef CONFIG_BL_SWITCHER
565static int bL_cpufreq_switcher_notifier(struct notifier_block *nfb,
566					unsigned long action, void *_arg)
567{
568	pr_debug("%s: action: %ld\n", __func__, action);
569
570	switch (action) {
571	case BL_NOTIFY_PRE_ENABLE:
572	case BL_NOTIFY_PRE_DISABLE:
573		cpufreq_unregister_driver(&bL_cpufreq_driver);
574		break;
575
576	case BL_NOTIFY_POST_ENABLE:
577		set_switching_enabled(true);
578		cpufreq_register_driver(&bL_cpufreq_driver);
579		break;
580
581	case BL_NOTIFY_POST_DISABLE:
582		set_switching_enabled(false);
583		cpufreq_register_driver(&bL_cpufreq_driver);
584		break;
585
586	default:
587		return NOTIFY_DONE;
588	}
589
590	return NOTIFY_OK;
591}
592
593static struct notifier_block bL_switcher_notifier = {
594	.notifier_call = bL_cpufreq_switcher_notifier,
595};
596
597static int __bLs_register_notifier(void)
598{
599	return bL_switcher_register_notifier(&bL_switcher_notifier);
600}
601
602static int __bLs_unregister_notifier(void)
603{
604	return bL_switcher_unregister_notifier(&bL_switcher_notifier);
605}
606#else
607static int __bLs_register_notifier(void) { return 0; }
608static int __bLs_unregister_notifier(void) { return 0; }
609#endif
610
611int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
612{
613	int ret, i;
614
615	if (arm_bL_ops) {
616		pr_debug("%s: Already registered: %s, exiting\n", __func__,
617				arm_bL_ops->name);
618		return -EBUSY;
619	}
620
621	if (!ops || !strlen(ops->name) || !ops->init_opp_table) {
622		pr_err("%s: Invalid arm_bL_ops, exiting\n", __func__);
623		return -ENODEV;
624	}
625
626	arm_bL_ops = ops;
627
628	set_switching_enabled(bL_switcher_get_enabled());
629
630	for (i = 0; i < MAX_CLUSTERS; i++)
631		mutex_init(&cluster_lock[i]);
632
633	ret = cpufreq_register_driver(&bL_cpufreq_driver);
634	if (ret) {
635		pr_info("%s: Failed registering platform driver: %s, err: %d\n",
636				__func__, ops->name, ret);
637		arm_bL_ops = NULL;
638	} else {
639		ret = __bLs_register_notifier();
640		if (ret) {
641			cpufreq_unregister_driver(&bL_cpufreq_driver);
642			arm_bL_ops = NULL;
643		} else {
644			pr_info("%s: Registered platform driver: %s\n",
645					__func__, ops->name);
646		}
647	}
648
649	bL_switcher_put_enabled();
650	return ret;
651}
652EXPORT_SYMBOL_GPL(bL_cpufreq_register);
653
654void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops)
655{
656	if (arm_bL_ops != ops) {
657		pr_err("%s: Registered with: %s, can't unregister, exiting\n",
658				__func__, arm_bL_ops->name);
659		return;
660	}
661
662	bL_switcher_get_enabled();
663	__bLs_unregister_notifier();
664	cpufreq_unregister_driver(&bL_cpufreq_driver);
665	bL_switcher_put_enabled();
666	pr_info("%s: Un-registered platform driver: %s\n", __func__,
667			arm_bL_ops->name);
668	arm_bL_ops = NULL;
669}
670EXPORT_SYMBOL_GPL(bL_cpufreq_unregister);
671
672MODULE_AUTHOR("Viresh Kumar <viresh.kumar@linaro.org>");
673MODULE_DESCRIPTION("Generic ARM big LITTLE cpufreq driver");
674MODULE_LICENSE("GPL v2");