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/export.h>
27#include <linux/module.h>
28#include <linux/mutex.h>
29#include <linux/of_platform.h>
30#include <linux/pm_opp.h>
31#include <linux/slab.h>
32#include <linux/topology.h>
33#include <linux/types.h>
34#include <asm/bL_switcher.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
44static bool bL_switching_enabled;
45#define is_bL_switching_enabled()	bL_switching_enabled
46#define set_switching_enabled(x)	(bL_switching_enabled = (x))
47#else
48#define is_bL_switching_enabled()	false
49#define set_switching_enabled(x)	do { } while (0)
50#endif
51
52#define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
53#define VIRT_FREQ(cluster, freq)    ((cluster == A7_CLUSTER) ? freq >> 1 : freq)
54
55static struct cpufreq_arm_bL_ops *arm_bL_ops;
56static struct clk *clk[MAX_CLUSTERS];
57static struct cpufreq_frequency_table *freq_table[MAX_CLUSTERS + 1];
58static atomic_t cluster_usage[MAX_CLUSTERS + 1];
59
60static unsigned int clk_big_min;	/* (Big) clock frequencies */
61static unsigned int clk_little_max;	/* Maximum clock frequency (Little) */
62
63static DEFINE_PER_CPU(unsigned int, physical_cluster);
64static DEFINE_PER_CPU(unsigned int, cpu_last_req_freq);
65
66static struct mutex cluster_lock[MAX_CLUSTERS];
67
68static inline int raw_cpu_to_cluster(int cpu)
69{
70	return topology_physical_package_id(cpu);
71}
72
73static inline int cpu_to_cluster(int cpu)
74{
75	return is_bL_switching_enabled() ?
76		MAX_CLUSTERS : raw_cpu_to_cluster(cpu);
77}
78
79static unsigned int find_cluster_maxfreq(int cluster)
80{
81	int j;
82	u32 max_freq = 0, cpu_freq;
83
84	for_each_online_cpu(j) {
85		cpu_freq = per_cpu(cpu_last_req_freq, j);
86
87		if ((cluster == per_cpu(physical_cluster, j)) &&
88				(max_freq < cpu_freq))
89			max_freq = cpu_freq;
90	}
91
92	pr_debug("%s: cluster: %d, max freq: %d\n", __func__, cluster,
93			max_freq);
94
95	return max_freq;
96}
97
98static unsigned int clk_get_cpu_rate(unsigned int cpu)
99{
100	u32 cur_cluster = per_cpu(physical_cluster, cpu);
101	u32 rate = clk_get_rate(clk[cur_cluster]) / 1000;
102
103	/* For switcher we use virtual A7 clock rates */
104	if (is_bL_switching_enabled())
105		rate = VIRT_FREQ(cur_cluster, rate);
106
107	pr_debug("%s: cpu: %d, cluster: %d, freq: %u\n", __func__, cpu,
108			cur_cluster, rate);
109
110	return rate;
111}
112
113static unsigned int bL_cpufreq_get_rate(unsigned int cpu)
114{
115	if (is_bL_switching_enabled()) {
116		pr_debug("%s: freq: %d\n", __func__, per_cpu(cpu_last_req_freq,
117					cpu));
118
119		return per_cpu(cpu_last_req_freq, cpu);
120	} else {
121		return clk_get_cpu_rate(cpu);
122	}
123}
124
125static unsigned int
126bL_cpufreq_set_rate(u32 cpu, u32 old_cluster, u32 new_cluster, u32 rate)
127{
128	u32 new_rate, prev_rate;
129	int ret;
130	bool bLs = is_bL_switching_enabled();
131
132	mutex_lock(&cluster_lock[new_cluster]);
133
134	if (bLs) {
135		prev_rate = per_cpu(cpu_last_req_freq, cpu);
136		per_cpu(cpu_last_req_freq, cpu) = rate;
137		per_cpu(physical_cluster, cpu) = new_cluster;
138
139		new_rate = find_cluster_maxfreq(new_cluster);
140		new_rate = ACTUAL_FREQ(new_cluster, new_rate);
141	} else {
142		new_rate = rate;
143	}
144
145	pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d, freq: %d\n",
146			__func__, cpu, old_cluster, new_cluster, new_rate);
147
148	ret = clk_set_rate(clk[new_cluster], new_rate * 1000);
149	if (WARN_ON(ret)) {
150		pr_err("clk_set_rate failed: %d, new cluster: %d\n", ret,
151				new_cluster);
152		if (bLs) {
153			per_cpu(cpu_last_req_freq, cpu) = prev_rate;
154			per_cpu(physical_cluster, cpu) = old_cluster;
155		}
156
157		mutex_unlock(&cluster_lock[new_cluster]);
158
159		return ret;
160	}
161
162	mutex_unlock(&cluster_lock[new_cluster]);
163
164	/* Recalc freq for old cluster when switching clusters */
165	if (old_cluster != new_cluster) {
166		pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d\n",
167				__func__, cpu, old_cluster, new_cluster);
168
169		/* Switch cluster */
170		bL_switch_request(cpu, new_cluster);
171
172		mutex_lock(&cluster_lock[old_cluster]);
173
174		/* Set freq of old cluster if there are cpus left on it */
175		new_rate = find_cluster_maxfreq(old_cluster);
176		new_rate = ACTUAL_FREQ(old_cluster, new_rate);
177
178		if (new_rate) {
179			pr_debug("%s: Updating rate of old cluster: %d, to freq: %d\n",
180					__func__, old_cluster, new_rate);
181
182			if (clk_set_rate(clk[old_cluster], new_rate * 1000))
183				pr_err("%s: clk_set_rate failed: %d, old cluster: %d\n",
184						__func__, ret, old_cluster);
185		}
186		mutex_unlock(&cluster_lock[old_cluster]);
187	}
188
189	return 0;
190}
191
192/* Set clock frequency */
193static int bL_cpufreq_set_target(struct cpufreq_policy *policy,
194		unsigned int index)
195{
196	u32 cpu = policy->cpu, cur_cluster, new_cluster, actual_cluster;
197	unsigned int freqs_new;
198
199	cur_cluster = cpu_to_cluster(cpu);
200	new_cluster = actual_cluster = per_cpu(physical_cluster, cpu);
201
202	freqs_new = freq_table[cur_cluster][index].frequency;
203
204	if (is_bL_switching_enabled()) {
205		if ((actual_cluster == A15_CLUSTER) &&
206				(freqs_new < clk_big_min)) {
207			new_cluster = A7_CLUSTER;
208		} else if ((actual_cluster == A7_CLUSTER) &&
209				(freqs_new > clk_little_max)) {
210			new_cluster = A15_CLUSTER;
211		}
212	}
213
214	return bL_cpufreq_set_rate(cpu, actual_cluster, new_cluster, freqs_new);
215}
216
217static inline u32 get_table_count(struct cpufreq_frequency_table *table)
218{
219	int count;
220
221	for (count = 0; table[count].frequency != CPUFREQ_TABLE_END; count++)
222		;
223
224	return count;
225}
226
227/* get the minimum frequency in the cpufreq_frequency_table */
228static inline u32 get_table_min(struct cpufreq_frequency_table *table)
229{
230	struct cpufreq_frequency_table *pos;
231	uint32_t min_freq = ~0;
232	cpufreq_for_each_entry(pos, table)
233		if (pos->frequency < min_freq)
234			min_freq = pos->frequency;
235	return min_freq;
236}
237
238/* get the maximum frequency in the cpufreq_frequency_table */
239static inline u32 get_table_max(struct cpufreq_frequency_table *table)
240{
241	struct cpufreq_frequency_table *pos;
242	uint32_t max_freq = 0;
243	cpufreq_for_each_entry(pos, table)
244		if (pos->frequency > max_freq)
245			max_freq = pos->frequency;
246	return max_freq;
247}
248
249static int merge_cluster_tables(void)
250{
251	int i, j, k = 0, count = 1;
252	struct cpufreq_frequency_table *table;
253
254	for (i = 0; i < MAX_CLUSTERS; i++)
255		count += get_table_count(freq_table[i]);
256
257	table = kzalloc(sizeof(*table) * count, GFP_KERNEL);
258	if (!table)
259		return -ENOMEM;
260
261	freq_table[MAX_CLUSTERS] = table;
262
263	/* Add in reverse order to get freqs in increasing order */
264	for (i = MAX_CLUSTERS - 1; i >= 0; i--) {
265		for (j = 0; freq_table[i][j].frequency != CPUFREQ_TABLE_END;
266				j++) {
267			table[k].frequency = VIRT_FREQ(i,
268					freq_table[i][j].frequency);
269			pr_debug("%s: index: %d, freq: %d\n", __func__, k,
270					table[k].frequency);
271			k++;
272		}
273	}
274
275	table[k].driver_data = k;
276	table[k].frequency = CPUFREQ_TABLE_END;
277
278	pr_debug("%s: End, table: %p, count: %d\n", __func__, table, k);
279
280	return 0;
281}
282
283static void _put_cluster_clk_and_freq_table(struct device *cpu_dev)
284{
285	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
286
287	if (!freq_table[cluster])
288		return;
289
290	clk_put(clk[cluster]);
291	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
292	if (arm_bL_ops->free_opp_table)
293		arm_bL_ops->free_opp_table(cpu_dev);
294	dev_dbg(cpu_dev, "%s: cluster: %d\n", __func__, cluster);
295}
296
297static void put_cluster_clk_and_freq_table(struct device *cpu_dev)
298{
299	u32 cluster = cpu_to_cluster(cpu_dev->id);
300	int i;
301
302	if (atomic_dec_return(&cluster_usage[cluster]))
303		return;
304
305	if (cluster < MAX_CLUSTERS)
306		return _put_cluster_clk_and_freq_table(cpu_dev);
307
308	for_each_present_cpu(i) {
309		struct device *cdev = get_cpu_device(i);
310		if (!cdev) {
311			pr_err("%s: failed to get cpu%d device\n", __func__, i);
312			return;
313		}
314
315		_put_cluster_clk_and_freq_table(cdev);
316	}
317
318	/* free virtual table */
319	kfree(freq_table[cluster]);
320}
321
322static int _get_cluster_clk_and_freq_table(struct device *cpu_dev)
323{
324	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
325	char name[14] = "cpu-cluster.";
326	int ret;
327
328	if (freq_table[cluster])
329		return 0;
330
331	ret = arm_bL_ops->init_opp_table(cpu_dev);
332	if (ret) {
333		dev_err(cpu_dev, "%s: init_opp_table failed, cpu: %d, err: %d\n",
334				__func__, cpu_dev->id, ret);
335		goto out;
336	}
337
338	ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster]);
339	if (ret) {
340		dev_err(cpu_dev, "%s: failed to init cpufreq table, cpu: %d, err: %d\n",
341				__func__, cpu_dev->id, ret);
342		goto free_opp_table;
343	}
344
345	name[12] = cluster + '0';
346	clk[cluster] = clk_get(cpu_dev, name);
347	if (!IS_ERR(clk[cluster])) {
348		dev_dbg(cpu_dev, "%s: clk: %p & freq table: %p, cluster: %d\n",
349				__func__, clk[cluster], freq_table[cluster],
350				cluster);
351		return 0;
352	}
353
354	dev_err(cpu_dev, "%s: Failed to get clk for cpu: %d, cluster: %d\n",
355			__func__, cpu_dev->id, cluster);
356	ret = PTR_ERR(clk[cluster]);
357	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
358
359free_opp_table:
360	if (arm_bL_ops->free_opp_table)
361		arm_bL_ops->free_opp_table(cpu_dev);
362out:
363	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n", __func__,
364			cluster);
365	return ret;
366}
367
368static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
369{
370	u32 cluster = cpu_to_cluster(cpu_dev->id);
371	int i, ret;
372
373	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
374		return 0;
375
376	if (cluster < MAX_CLUSTERS) {
377		ret = _get_cluster_clk_and_freq_table(cpu_dev);
378		if (ret)
379			atomic_dec(&cluster_usage[cluster]);
380		return ret;
381	}
382
383	/*
384	 * Get data for all clusters and fill virtual cluster with a merge of
385	 * both
386	 */
387	for_each_present_cpu(i) {
388		struct device *cdev = get_cpu_device(i);
389		if (!cdev) {
390			pr_err("%s: failed to get cpu%d device\n", __func__, i);
391			return -ENODEV;
392		}
393
394		ret = _get_cluster_clk_and_freq_table(cdev);
395		if (ret)
396			goto put_clusters;
397	}
398
399	ret = merge_cluster_tables();
400	if (ret)
401		goto put_clusters;
402
403	/* Assuming 2 cluster, set clk_big_min and clk_little_max */
404	clk_big_min = get_table_min(freq_table[0]);
405	clk_little_max = VIRT_FREQ(1, get_table_max(freq_table[1]));
406
407	pr_debug("%s: cluster: %d, clk_big_min: %d, clk_little_max: %d\n",
408			__func__, cluster, clk_big_min, clk_little_max);
409
410	return 0;
411
412put_clusters:
413	for_each_present_cpu(i) {
414		struct device *cdev = get_cpu_device(i);
415		if (!cdev) {
416			pr_err("%s: failed to get cpu%d device\n", __func__, i);
417			return -ENODEV;
418		}
419
420		_put_cluster_clk_and_freq_table(cdev);
421	}
422
423	atomic_dec(&cluster_usage[cluster]);
424
425	return ret;
426}
427
428/* Per-CPU initialization */
429static int bL_cpufreq_init(struct cpufreq_policy *policy)
430{
431	u32 cur_cluster = cpu_to_cluster(policy->cpu);
432	struct device *cpu_dev;
433	int ret;
434
435	cpu_dev = get_cpu_device(policy->cpu);
436	if (!cpu_dev) {
437		pr_err("%s: failed to get cpu%d device\n", __func__,
438				policy->cpu);
439		return -ENODEV;
440	}
441
442	ret = get_cluster_clk_and_freq_table(cpu_dev);
443	if (ret)
444		return ret;
445
446	ret = cpufreq_table_validate_and_show(policy, freq_table[cur_cluster]);
447	if (ret) {
448		dev_err(cpu_dev, "CPU %d, cluster: %d invalid freq table\n",
449				policy->cpu, cur_cluster);
450		put_cluster_clk_and_freq_table(cpu_dev);
451		return ret;
452	}
453
454	if (cur_cluster < MAX_CLUSTERS) {
455		int cpu;
456
457		cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
458
459		for_each_cpu(cpu, policy->cpus)
460			per_cpu(physical_cluster, cpu) = cur_cluster;
461	} else {
462		/* Assumption: during init, we are always running on A15 */
463		per_cpu(physical_cluster, policy->cpu) = A15_CLUSTER;
464	}
465
466	if (arm_bL_ops->get_transition_latency)
467		policy->cpuinfo.transition_latency =
468			arm_bL_ops->get_transition_latency(cpu_dev);
469	else
470		policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
471
472	if (is_bL_switching_enabled())
473		per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate(policy->cpu);
474
475	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
476	return 0;
477}
478
479static int bL_cpufreq_exit(struct cpufreq_policy *policy)
480{
481	struct device *cpu_dev;
482
483	cpu_dev = get_cpu_device(policy->cpu);
484	if (!cpu_dev) {
485		pr_err("%s: failed to get cpu%d device\n", __func__,
486				policy->cpu);
487		return -ENODEV;
488	}
489
490	put_cluster_clk_and_freq_table(cpu_dev);
491	dev_dbg(cpu_dev, "%s: Exited, cpu: %d\n", __func__, policy->cpu);
492
493	return 0;
494}
495
496static struct cpufreq_driver bL_cpufreq_driver = {
497	.name			= "arm-big-little",
498	.flags			= CPUFREQ_STICKY |
499					CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
500					CPUFREQ_NEED_INITIAL_FREQ_CHECK,
501	.verify			= cpufreq_generic_frequency_table_verify,
502	.target_index		= bL_cpufreq_set_target,
503	.get			= bL_cpufreq_get_rate,
504	.init			= bL_cpufreq_init,
505	.exit			= bL_cpufreq_exit,
506	.attr			= cpufreq_generic_attr,
507};
508
509static int bL_cpufreq_switcher_notifier(struct notifier_block *nfb,
510					unsigned long action, void *_arg)
511{
512	pr_debug("%s: action: %ld\n", __func__, action);
513
514	switch (action) {
515	case BL_NOTIFY_PRE_ENABLE:
516	case BL_NOTIFY_PRE_DISABLE:
517		cpufreq_unregister_driver(&bL_cpufreq_driver);
518		break;
519
520	case BL_NOTIFY_POST_ENABLE:
521		set_switching_enabled(true);
522		cpufreq_register_driver(&bL_cpufreq_driver);
523		break;
524
525	case BL_NOTIFY_POST_DISABLE:
526		set_switching_enabled(false);
527		cpufreq_register_driver(&bL_cpufreq_driver);
528		break;
529
530	default:
531		return NOTIFY_DONE;
532	}
533
534	return NOTIFY_OK;
535}
536
537static struct notifier_block bL_switcher_notifier = {
538	.notifier_call = bL_cpufreq_switcher_notifier,
539};
540
541int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
542{
543	int ret, i;
544
545	if (arm_bL_ops) {
546		pr_debug("%s: Already registered: %s, exiting\n", __func__,
547				arm_bL_ops->name);
548		return -EBUSY;
549	}
550
551	if (!ops || !strlen(ops->name) || !ops->init_opp_table) {
552		pr_err("%s: Invalid arm_bL_ops, exiting\n", __func__);
553		return -ENODEV;
554	}
555
556	arm_bL_ops = ops;
557
558	ret = bL_switcher_get_enabled();
559	set_switching_enabled(ret);
560
561	for (i = 0; i < MAX_CLUSTERS; i++)
562		mutex_init(&cluster_lock[i]);
563
564	ret = cpufreq_register_driver(&bL_cpufreq_driver);
565	if (ret) {
566		pr_info("%s: Failed registering platform driver: %s, err: %d\n",
567				__func__, ops->name, ret);
568		arm_bL_ops = NULL;
569	} else {
570		ret = bL_switcher_register_notifier(&bL_switcher_notifier);
571		if (ret) {
572			cpufreq_unregister_driver(&bL_cpufreq_driver);
573			arm_bL_ops = NULL;
574		} else {
575			pr_info("%s: Registered platform driver: %s\n",
576					__func__, ops->name);
577		}
578	}
579
580	bL_switcher_put_enabled();
581	return ret;
582}
583EXPORT_SYMBOL_GPL(bL_cpufreq_register);
584
585void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops)
586{
587	if (arm_bL_ops != ops) {
588		pr_err("%s: Registered with: %s, can't unregister, exiting\n",
589				__func__, arm_bL_ops->name);
590		return;
591	}
592
593	bL_switcher_get_enabled();
594	bL_switcher_unregister_notifier(&bL_switcher_notifier);
595	cpufreq_unregister_driver(&bL_cpufreq_driver);
596	bL_switcher_put_enabled();
597	pr_info("%s: Un-registered platform driver: %s\n", __func__,
598			arm_bL_ops->name);
599	arm_bL_ops = NULL;
600}
601EXPORT_SYMBOL_GPL(bL_cpufreq_unregister);
602
603MODULE_AUTHOR("Viresh Kumar <viresh.kumar@linaro.org>");
604MODULE_DESCRIPTION("Generic ARM big LITTLE cpufreq driver");
605MODULE_LICENSE("GPL v2");
606