1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Versatile Express SPC CPUFreq Interface driver
  4 *
  5 * Copyright (C) 2013 - 2019 ARM Ltd.
  6 * Sudeep Holla <sudeep.holla@arm.com>
  7 *
  8 * Copyright (C) 2013 Linaro.
  9 * Viresh Kumar <viresh.kumar@linaro.org>
 10 */
 11
 12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 13
 14#include <linux/clk.h>
 15#include <linux/cpu.h>
 16#include <linux/cpufreq.h>
 17#include <linux/cpumask.h>
 18#include <linux/cpu_cooling.h>
 19#include <linux/device.h>
 20#include <linux/module.h>
 21#include <linux/mutex.h>
 22#include <linux/of_platform.h>
 23#include <linux/platform_device.h>
 24#include <linux/pm_opp.h>
 25#include <linux/slab.h>
 26#include <linux/topology.h>
 27#include <linux/types.h>
 28
 29/* Currently we support only two clusters */
 30#define A15_CLUSTER	0
 31#define A7_CLUSTER	1
 32#define MAX_CLUSTERS	2
 33
 34#ifdef CONFIG_BL_SWITCHER
 35#include <asm/bL_switcher.h>
 36static bool bL_switching_enabled;
 37#define is_bL_switching_enabled()	bL_switching_enabled
 38#define set_switching_enabled(x)	(bL_switching_enabled = (x))
 39#else
 40#define is_bL_switching_enabled()	false
 41#define set_switching_enabled(x)	do { } while (0)
 42#define bL_switch_request(...)		do { } while (0)
 43#define bL_switcher_put_enabled()	do { } while (0)
 44#define bL_switcher_get_enabled()	do { } while (0)
 45#endif
 46
 47#define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
 48#define VIRT_FREQ(cluster, freq)    ((cluster == A7_CLUSTER) ? freq >> 1 : freq)
 49
 50static struct thermal_cooling_device *cdev[MAX_CLUSTERS];
 51static struct clk *clk[MAX_CLUSTERS];
 52static struct cpufreq_frequency_table *freq_table[MAX_CLUSTERS + 1];
 53static atomic_t cluster_usage[MAX_CLUSTERS + 1];
 54
 55static unsigned int clk_big_min;	/* (Big) clock frequencies */
 56static unsigned int clk_little_max;	/* Maximum clock frequency (Little) */
 57
 58static DEFINE_PER_CPU(unsigned int, physical_cluster);
 59static DEFINE_PER_CPU(unsigned int, cpu_last_req_freq);
 60
 61static struct mutex cluster_lock[MAX_CLUSTERS];
 62
 63static inline int raw_cpu_to_cluster(int cpu)
 64{
 65	return topology_physical_package_id(cpu);
 66}
 67
 68static inline int cpu_to_cluster(int cpu)
 69{
 70	return is_bL_switching_enabled() ?
 71		MAX_CLUSTERS : raw_cpu_to_cluster(cpu);
 72}
 73
 74static unsigned int find_cluster_maxfreq(int cluster)
 75{
 76	int j;
 77	u32 max_freq = 0, cpu_freq;
 78
 79	for_each_online_cpu(j) {
 80		cpu_freq = per_cpu(cpu_last_req_freq, j);
 81
 82		if (cluster == per_cpu(physical_cluster, j) &&
 83		    max_freq < cpu_freq)
 84			max_freq = cpu_freq;
 85	}
 86
 87	return max_freq;
 88}
 89
 90static unsigned int clk_get_cpu_rate(unsigned int cpu)
 91{
 92	u32 cur_cluster = per_cpu(physical_cluster, cpu);
 93	u32 rate = clk_get_rate(clk[cur_cluster]) / 1000;
 94
 95	/* For switcher we use virtual A7 clock rates */
 96	if (is_bL_switching_enabled())
 97		rate = VIRT_FREQ(cur_cluster, rate);
 98
 99	return rate;
100}
101
102static unsigned int ve_spc_cpufreq_get_rate(unsigned int cpu)
103{
104	if (is_bL_switching_enabled())
105		return per_cpu(cpu_last_req_freq, cpu);
106	else
107		return clk_get_cpu_rate(cpu);
108}
109
110static unsigned int
111ve_spc_cpufreq_set_rate(u32 cpu, u32 old_cluster, u32 new_cluster, u32 rate)
112{
113	u32 new_rate, prev_rate;
114	int ret;
115	bool bLs = is_bL_switching_enabled();
116
117	mutex_lock(&cluster_lock[new_cluster]);
118
119	if (bLs) {
120		prev_rate = per_cpu(cpu_last_req_freq, cpu);
121		per_cpu(cpu_last_req_freq, cpu) = rate;
122		per_cpu(physical_cluster, cpu) = new_cluster;
123
124		new_rate = find_cluster_maxfreq(new_cluster);
125		new_rate = ACTUAL_FREQ(new_cluster, new_rate);
126	} else {
127		new_rate = rate;
128	}
129
130	ret = clk_set_rate(clk[new_cluster], new_rate * 1000);
131	if (!ret) {
132		/*
133		 * FIXME: clk_set_rate hasn't returned an error here however it
134		 * may be that clk_change_rate failed due to hardware or
135		 * firmware issues and wasn't able to report that due to the
136		 * current design of the clk core layer. To work around this
137		 * problem we will read back the clock rate and check it is
138		 * correct. This needs to be removed once clk core is fixed.
139		 */
140		if (clk_get_rate(clk[new_cluster]) != new_rate * 1000)
141			ret = -EIO;
142	}
143
144	if (WARN_ON(ret)) {
145		if (bLs) {
146			per_cpu(cpu_last_req_freq, cpu) = prev_rate;
147			per_cpu(physical_cluster, cpu) = old_cluster;
148		}
149
150		mutex_unlock(&cluster_lock[new_cluster]);
151
152		return ret;
153	}
154
155	mutex_unlock(&cluster_lock[new_cluster]);
156
157	/* Recalc freq for old cluster when switching clusters */
158	if (old_cluster != new_cluster) {
159		/* Switch cluster */
160		bL_switch_request(cpu, new_cluster);
161
162		mutex_lock(&cluster_lock[old_cluster]);
163
164		/* Set freq of old cluster if there are cpus left on it */
165		new_rate = find_cluster_maxfreq(old_cluster);
166		new_rate = ACTUAL_FREQ(old_cluster, new_rate);
167
168		if (new_rate &&
169		    clk_set_rate(clk[old_cluster], new_rate * 1000)) {
170			pr_err("%s: clk_set_rate failed: %d, old cluster: %d\n",
171			       __func__, ret, old_cluster);
172		}
173		mutex_unlock(&cluster_lock[old_cluster]);
174	}
175
176	return 0;
177}
178
179/* Set clock frequency */
180static int ve_spc_cpufreq_set_target(struct cpufreq_policy *policy,
181				     unsigned int index)
182{
183	u32 cpu = policy->cpu, cur_cluster, new_cluster, actual_cluster;
184	unsigned int freqs_new;
185
186	cur_cluster = cpu_to_cluster(cpu);
187	new_cluster = actual_cluster = per_cpu(physical_cluster, cpu);
188
189	freqs_new = freq_table[cur_cluster][index].frequency;
190
191	if (is_bL_switching_enabled()) {
192		if (actual_cluster == A15_CLUSTER && freqs_new < clk_big_min)
193			new_cluster = A7_CLUSTER;
194		else if (actual_cluster == A7_CLUSTER &&
195			 freqs_new > clk_little_max)
196			new_cluster = A15_CLUSTER;
197	}
198
199	return ve_spc_cpufreq_set_rate(cpu, actual_cluster, new_cluster,
200				       freqs_new);
201}
202
203static inline u32 get_table_count(struct cpufreq_frequency_table *table)
204{
205	int count;
206
207	for (count = 0; table[count].frequency != CPUFREQ_TABLE_END; count++)
208		;
209
210	return count;
211}
212
213/* get the minimum frequency in the cpufreq_frequency_table */
214static inline u32 get_table_min(struct cpufreq_frequency_table *table)
215{
216	struct cpufreq_frequency_table *pos;
217	u32 min_freq = ~0;
218
219	cpufreq_for_each_entry(pos, table)
220		if (pos->frequency < min_freq)
221			min_freq = pos->frequency;
222	return min_freq;
223}
224
225/* get the maximum frequency in the cpufreq_frequency_table */
226static inline u32 get_table_max(struct cpufreq_frequency_table *table)
227{
228	struct cpufreq_frequency_table *pos;
229	u32 max_freq = 0;
230
231	cpufreq_for_each_entry(pos, table)
232		if (pos->frequency > max_freq)
233			max_freq = pos->frequency;
234	return max_freq;
235}
236
237static bool search_frequency(struct cpufreq_frequency_table *table, int size,
238			     unsigned int freq)
239{
240	int count;
241
242	for (count = 0; count < size; count++) {
243		if (table[count].frequency == freq)
244			return true;
245	}
246
247	return false;
248}
249
250static int merge_cluster_tables(void)
251{
252	int i, j, k = 0, count = 1;
253	struct cpufreq_frequency_table *table;
254
255	for (i = 0; i < MAX_CLUSTERS; i++)
256		count += get_table_count(freq_table[i]);
257
258	table = kcalloc(count, sizeof(*table), GFP_KERNEL);
259	if (!table)
260		return -ENOMEM;
261
262	freq_table[MAX_CLUSTERS] = table;
263
264	/* Add in reverse order to get freqs in increasing order */
265	for (i = MAX_CLUSTERS - 1; i >= 0; i--, count = k) {
266		for (j = 0; freq_table[i][j].frequency != CPUFREQ_TABLE_END;
267		     j++) {
268			if (i == A15_CLUSTER &&
269			    search_frequency(table, count, freq_table[i][j].frequency))
270				continue; /* skip duplicates */
271			table[k++].frequency =
272				VIRT_FREQ(i, freq_table[i][j].frequency);
273		}
274	}
275
276	table[k].driver_data = k;
277	table[k].frequency = CPUFREQ_TABLE_END;
278
279	return 0;
280}
281
282static void _put_cluster_clk_and_freq_table(struct device *cpu_dev,
283					    const struct cpumask *cpumask)
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}
293
294static void put_cluster_clk_and_freq_table(struct device *cpu_dev,
295					   const struct cpumask *cpumask)
296{
297	u32 cluster = cpu_to_cluster(cpu_dev->id);
298	int i;
299
300	if (atomic_dec_return(&cluster_usage[cluster]))
301		return;
302
303	if (cluster < MAX_CLUSTERS)
304		return _put_cluster_clk_and_freq_table(cpu_dev, cpumask);
305
306	for_each_present_cpu(i) {
307		struct device *cdev = get_cpu_device(i);
308
309		if (!cdev)
310			return;
311
312		_put_cluster_clk_and_freq_table(cdev, cpumask);
313	}
314
315	/* free virtual table */
316	kfree(freq_table[cluster]);
317}
318
319static int _get_cluster_clk_and_freq_table(struct device *cpu_dev,
320					   const struct cpumask *cpumask)
321{
322	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
323	int ret;
324
325	if (freq_table[cluster])
326		return 0;
327
328	/*
329	 * platform specific SPC code must initialise the opp table
330	 * so just check if the OPP count is non-zero
331	 */
332	ret = dev_pm_opp_get_opp_count(cpu_dev) <= 0;
333	if (ret)
334		goto out;
335
336	ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster]);
337	if (ret)
338		goto out;
339
340	clk[cluster] = clk_get(cpu_dev, NULL);
341	if (!IS_ERR(clk[cluster]))
342		return 0;
343
344	dev_err(cpu_dev, "%s: Failed to get clk for cpu: %d, cluster: %d\n",
345		__func__, cpu_dev->id, cluster);
346	ret = PTR_ERR(clk[cluster]);
347	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
348
349out:
350	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n", __func__,
351		cluster);
352	return ret;
353}
354
355static int get_cluster_clk_and_freq_table(struct device *cpu_dev,
356					  const struct cpumask *cpumask)
357{
358	u32 cluster = cpu_to_cluster(cpu_dev->id);
359	int i, ret;
360
361	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
362		return 0;
363
364	if (cluster < MAX_CLUSTERS) {
365		ret = _get_cluster_clk_and_freq_table(cpu_dev, cpumask);
366		if (ret)
367			atomic_dec(&cluster_usage[cluster]);
368		return ret;
369	}
370
371	/*
372	 * Get data for all clusters and fill virtual cluster with a merge of
373	 * both
374	 */
375	for_each_present_cpu(i) {
376		struct device *cdev = get_cpu_device(i);
377
378		if (!cdev)
379			return -ENODEV;
380
381		ret = _get_cluster_clk_and_freq_table(cdev, cpumask);
382		if (ret)
383			goto put_clusters;
384	}
385
386	ret = merge_cluster_tables();
387	if (ret)
388		goto put_clusters;
389
390	/* Assuming 2 cluster, set clk_big_min and clk_little_max */
391	clk_big_min = get_table_min(freq_table[A15_CLUSTER]);
392	clk_little_max = VIRT_FREQ(A7_CLUSTER,
393				   get_table_max(freq_table[A7_CLUSTER]));
394
395	return 0;
396
397put_clusters:
398	for_each_present_cpu(i) {
399		struct device *cdev = get_cpu_device(i);
400
401		if (!cdev)
402			return -ENODEV;
403
404		_put_cluster_clk_and_freq_table(cdev, cpumask);
405	}
406
407	atomic_dec(&cluster_usage[cluster]);
408
409	return ret;
410}
411
412/* Per-CPU initialization */
413static int ve_spc_cpufreq_init(struct cpufreq_policy *policy)
414{
415	u32 cur_cluster = cpu_to_cluster(policy->cpu);
416	struct device *cpu_dev;
417	int ret;
418
419	cpu_dev = get_cpu_device(policy->cpu);
420	if (!cpu_dev) {
421		pr_err("%s: failed to get cpu%d device\n", __func__,
422		       policy->cpu);
423		return -ENODEV;
424	}
425
426	if (cur_cluster < MAX_CLUSTERS) {
427		int cpu;
428
429		dev_pm_opp_get_sharing_cpus(cpu_dev, policy->cpus);
430
431		for_each_cpu(cpu, policy->cpus)
432			per_cpu(physical_cluster, cpu) = cur_cluster;
433	} else {
434		/* Assumption: during init, we are always running on A15 */
435		per_cpu(physical_cluster, policy->cpu) = A15_CLUSTER;
436	}
437
438	ret = get_cluster_clk_and_freq_table(cpu_dev, policy->cpus);
439	if (ret)
440		return ret;
441
442	policy->freq_table = freq_table[cur_cluster];
443	policy->cpuinfo.transition_latency = 1000000; /* 1 ms */
444
445	dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
446
447	if (is_bL_switching_enabled())
448		per_cpu(cpu_last_req_freq, policy->cpu) =
449						clk_get_cpu_rate(policy->cpu);
450
451	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
452	return 0;
453}
454
455static int ve_spc_cpufreq_exit(struct cpufreq_policy *policy)
456{
457	struct device *cpu_dev;
458	int cur_cluster = cpu_to_cluster(policy->cpu);
459
460	if (cur_cluster < MAX_CLUSTERS) {
461		cpufreq_cooling_unregister(cdev[cur_cluster]);
462		cdev[cur_cluster] = NULL;
463	}
464
465	cpu_dev = get_cpu_device(policy->cpu);
466	if (!cpu_dev) {
467		pr_err("%s: failed to get cpu%d device\n", __func__,
468		       policy->cpu);
469		return -ENODEV;
470	}
471
472	put_cluster_clk_and_freq_table(cpu_dev, policy->related_cpus);
473	return 0;
474}
475
476static void ve_spc_cpufreq_ready(struct cpufreq_policy *policy)
477{
478	int cur_cluster = cpu_to_cluster(policy->cpu);
479
480	/* Do not register a cpu_cooling device if we are in IKS mode */
481	if (cur_cluster >= MAX_CLUSTERS)
482		return;
483
484	cdev[cur_cluster] = of_cpufreq_cooling_register(policy);
485}
486
487static struct cpufreq_driver ve_spc_cpufreq_driver = {
488	.name			= "vexpress-spc",
489	.flags			= CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
490					CPUFREQ_NEED_INITIAL_FREQ_CHECK,
491	.verify			= cpufreq_generic_frequency_table_verify,
492	.target_index		= ve_spc_cpufreq_set_target,
493	.get			= ve_spc_cpufreq_get_rate,
494	.init			= ve_spc_cpufreq_init,
495	.exit			= ve_spc_cpufreq_exit,
496	.ready			= ve_spc_cpufreq_ready,
497	.attr			= cpufreq_generic_attr,
498};
499
500#ifdef CONFIG_BL_SWITCHER
501static int bL_cpufreq_switcher_notifier(struct notifier_block *nfb,
502					unsigned long action, void *_arg)
503{
504	pr_debug("%s: action: %ld\n", __func__, action);
505
506	switch (action) {
507	case BL_NOTIFY_PRE_ENABLE:
508	case BL_NOTIFY_PRE_DISABLE:
509		cpufreq_unregister_driver(&ve_spc_cpufreq_driver);
510		break;
511
512	case BL_NOTIFY_POST_ENABLE:
513		set_switching_enabled(true);
514		cpufreq_register_driver(&ve_spc_cpufreq_driver);
515		break;
516
517	case BL_NOTIFY_POST_DISABLE:
518		set_switching_enabled(false);
519		cpufreq_register_driver(&ve_spc_cpufreq_driver);
520		break;
521
522	default:
523		return NOTIFY_DONE;
524	}
525
526	return NOTIFY_OK;
527}
528
529static struct notifier_block bL_switcher_notifier = {
530	.notifier_call = bL_cpufreq_switcher_notifier,
531};
532
533static int __bLs_register_notifier(void)
534{
535	return bL_switcher_register_notifier(&bL_switcher_notifier);
536}
537
538static int __bLs_unregister_notifier(void)
539{
540	return bL_switcher_unregister_notifier(&bL_switcher_notifier);
541}
542#else
543static int __bLs_register_notifier(void) { return 0; }
544static int __bLs_unregister_notifier(void) { return 0; }
545#endif
546
547static int ve_spc_cpufreq_probe(struct platform_device *pdev)
548{
549	int ret, i;
550
551	set_switching_enabled(bL_switcher_get_enabled());
552
553	for (i = 0; i < MAX_CLUSTERS; i++)
554		mutex_init(&cluster_lock[i]);
555
556	ret = cpufreq_register_driver(&ve_spc_cpufreq_driver);
557	if (ret) {
558		pr_info("%s: Failed registering platform driver: %s, err: %d\n",
559			__func__, ve_spc_cpufreq_driver.name, ret);
560	} else {
561		ret = __bLs_register_notifier();
562		if (ret)
563			cpufreq_unregister_driver(&ve_spc_cpufreq_driver);
564		else
565			pr_info("%s: Registered platform driver: %s\n",
566				__func__, ve_spc_cpufreq_driver.name);
567	}
568
569	bL_switcher_put_enabled();
570	return ret;
571}
572
573static int ve_spc_cpufreq_remove(struct platform_device *pdev)
574{
575	bL_switcher_get_enabled();
576	__bLs_unregister_notifier();
577	cpufreq_unregister_driver(&ve_spc_cpufreq_driver);
578	bL_switcher_put_enabled();
579	pr_info("%s: Un-registered platform driver: %s\n", __func__,
580		ve_spc_cpufreq_driver.name);
581	return 0;
582}
583
584static struct platform_driver ve_spc_cpufreq_platdrv = {
585	.driver = {
586		.name	= "vexpress-spc-cpufreq",
587	},
588	.probe		= ve_spc_cpufreq_probe,
589	.remove		= ve_spc_cpufreq_remove,
590};
591module_platform_driver(ve_spc_cpufreq_platdrv);
592
593MODULE_ALIAS("platform:vexpress-spc-cpufreq");
594MODULE_AUTHOR("Viresh Kumar <viresh.kumar@linaro.org>");
595MODULE_AUTHOR("Sudeep Holla <sudeep.holla@arm.com>");
596MODULE_DESCRIPTION("Vexpress SPC ARM big LITTLE cpufreq driver");
597MODULE_LICENSE("GPL v2");